Mediator Complex Subunits MED2, MED5, MED16, and MED23 Genetically Interact in the Regulation of Phenylpropanoid Biosynthesis

The effects of a semidominant MED5b mutant on phenylpropanoid metabolism and/or growth are suppressed by mutation of MED2, MED16, or MED23.

Abstract

The phenylpropanoid pathway is a major global carbon sink and is important for plant fitness and the engineering of bioenergy feedstocks. In Arabidopsis thaliana, disruption of two subunits of the transcriptional regulatory Mediator complex, MED5a and MED5b, results in an increase in phenylpropanoid accumulation. By contrast, the semidominant MED5b mutation reduced epidermal fluorescence4-3 (ref4-3) results in dwarfism and constitutively repressed phenylpropanoid accumulation. Here, we report the results of a forward genetic screen for suppressors of ref4-3. We identified 13 independent lines that restore growth and/or phenylpropanoid accumulation in the ref4-3 background. Two of the suppressors restore growth without restoring soluble phenylpropanoid accumulation, indicating that the growth and metabolic phenotypes of the ref4-3 mutant can be genetically disentangled. Whole-genome sequencing revealed that all but one of the suppressors carry mutations in MED5b or other Mediator subunits. RNA-seq analysis showed that the ref4-3 mutation causes widespread changes in gene expression, including the upregulation of negative regulators of the phenylpropanoid pathway, and that the suppressors reverse many of these changes. Together, our data highlight the interdependence of individual Mediator subunits and provide greater insight into the transcriptional regulation of phenylpropanoid biosynthesis by the Mediator complex.

INTRODUCTION

The Mediator complex is a hub for transcription regulation (Malik and Roeder, 2010) that is conserved across eukaryotes (Bourbon, 2008; Dolan and Chapple, 2017). The complex comprises some 25 to 34 subunits that are structurally divided into four modules known as the head, middle, tail, and kinase modules (Dotson et al., 2000). Mediator is recruited to promoters through interaction with enhancer-bound transcription factors and in turn facilitates the assembly of the preinitiation complex (Meyer et al., 2010; Ebmeier and Taatjes, 2010). Studies of Arabidopsis thaliana Mediator subunits have demonstrated the importance of individual MED subunits for proper regulation of numerous aspects of plant development, metabolism, and response to pathogens and environmental perturbations (reviewed in Samanta and Thakur, 2015; Yang et al., 2016). Most studies of plant Mediator have focused on one or two Mediator (MED) subunits and their roles in gene regulation; however, MED subunits do not function in isolation and greater understanding of Mediator function requires additional knowledge of the interdependencies between complex subunits. At present, the relative positions of the Arabidopsis MED subunits have been provisionally assigned based on the positions of their homologs in yeast and human complexes; however, to date, there have been no structural determinations of the Arabidopsis complex, and as such, it is not clear how perturbation of a given subunit might affect the function of other subunits or the complex as a whole.

Plants produce many secondary metabolites from the amino acid phenylalanine that are important to plant fitness (Douglas, 1996; Vogt, 2010). These compounds are collectively referred to as phenylpropanoids and have a variety of specialized roles. Lignin, for example, is deposited in the secondary cell wall where it provides the mechanical strength and hydrophobicity to tracheary elements essential for water transport (Cooper-Driver, 2001). Sinapoylmalate, a soluble phenylpropanoid, accumulates in the upper epidermis of Arabidopsis leaves where it functions as a UV protectant (Landry et al., 1995). Flavonoids, another large class of soluble phenylpropanoids, include pigmented compounds that attract pollinators (Grotewold, 2006) and that accumulate in response to many environmental stresses, protecting cells from oxidative damage (Winkel-Shirley, 2002; Nakabayashi et al., 2014).

Phenylpropanoid biosynthesis is a significant metabolic commitment for plants and a major global carbon sink (Field et al., 1998) that could be tapped for bioengineering of renewable fuels or chemical feedstocks. Genetic manipulation of lignin content and composition have variable and unexpected effects on plant fitness, metabolite accumulation, and biomass production (Bonawitz and Chapple, 2010; Vanholme et al., 2012). The unpredictability of the phenotypic outcomes of these genetic changes demonstrates our incomplete understanding of the regulation of phenylpropanoid biosynthesis. Other metabolic pathways (Hemm et al., 2003; Kim et al., 2015), as well as developmental and stress response systems (Douglas et al., 1992; Taylor-Teeples et al., 2015), are known to impinge on phenylpropanoid biosynthesis, making for a complex regulatory network with many potential interactions and outcomes. Numerous transcription factors that regulate phenylpropanoid biosynthesis and secondary cell wall formation have been identified (Zhong et al., 2008; Taylor-Teeples et al., 2015) and systematic studies of the metabolic and transcriptional consequences of perturbations in the lignin pathway (Vanholme et al., 2012) have been undertaken; however, it is not clear how these signals are integrated in vivo.

In Arabidopsis, two MED5 homologs, MED5a and MED5b, are required for phenylpropanoid homeostasis (Bonawitz et al., 2012). In a screen for plants defective in phenylpropanoid biosynthesis, we identified a semidominant mutant of MED5b, which we named reduced epidermal fluorescence 4-3 (ref4-3). The ref4-3 mutant is dwarf, has reduced lignin content, and accumulates low levels of soluble phenylpropanoids (Ruegger and Chapple, 2001; Stout et al., 2008). Conversely, plants that are null for REF4/MED5b and RFR1/MED5a (med5ab) exhibit normal growth and accumulate increased levels of phenylpropanoids (Stout et al., 2008; Bonawitz et al., 2012). Gene expression analysis showed that transcripts of the gene encoding the phenylpropanoid biosynthetic enzyme PHENYLALANINE AMMONIA LYASE1 are increased in med5ab double mutants and decreased in ref4-3 mutants, in accordance with their phenylpropanoid content (Bonawitz et al., 2012).

More recently, we showed that MED5 is required for feedback and crosstalk that arise from genetic perturbations in phenylpropanoid (Bonawitz et al., 2014; Anderson et al., 2015) or glucosinolate metabolism (Kim et al., 2015). These observations suggest that MED5 alters phenylpropanoid biosynthesis in response to metabolic changes and may function as a regulatory node for the pathway. Many of the components of these signaling pathways are still unknown. For example, we neither know the molecular identities of signals to which MED5 responds nor the breadth of its downstream targets. The mutant protein encoded by the ref4-3 allele of MED5b constitutively downregulates phenylpropanoid gene expression but presumably requires other proteins to do so. To identify genes required for ref4-3 function, or for the phenotypes that the mutant exhibits, we performed a genetic screen for mutants that suppress ref4-3. The molecular identities of these suppressor alleles were determined by whole-genome sequencing and confirmed by complementation tests with additional alleles. We show here that all but one of the suppressors of the ref4-3 allele of Med5b that we isolated harbor intragenic mutations in MED5b or mutations in the MED2, MED16, and MED23 subunits of Mediator. Transcriptomic analysis of the mutants suggests that ref4-3 represses phenylpropanoid biosynthesis by upregulating negative regulators of the pathway. In addition, two of the suppressors we identified restore growth without restoring soluble phenylpropanoid accumulation, demonstrating that the two phenotypes can be uncoupled. Together, these genetic data identify important functional relationships between subunits of the Mediator complex in the transcriptional control of gene expression and its consequences for metabolism.

RESULTS

Suppressors of ref4-3 Differentially Restore Growth and Sinapoylmalate Accumulation

We previously reported on a ref4-3 suppressor screen that was used to aid in the identification of the REF4 gene (Stout et al., 2008). Approximately 75,000 ref4-3 seeds were treated with EMS and the M2 generation was screened for restored growth or UV fluorescence when compared with ref4-3. From this M2 population, we identified 14 ref4-3 suppressors (Figure 1). Quantification of sinapoylmalate in the suppressors showed that their sinapoylmalate content ranges from the levels found in ref4-3 to the wild type (Figure 1A). The suppressors also show a variety of rosette sizes and morphologies, with some being compact and small with rounded leaves, similar to ref4-3, some intermediate in growth, and others being indistinguishable from the wild type (Figure 1B). Notably, two of the ref four suppressors (rfs) had low sinapoylmalate contents and were substantially restored in their growth (rfs7-4 and rfs33-2), suggesting that the growth and soluble phenylpropanoid phenotypes of ref4-3 may come about by different mechanisms.

Figure 1.

The ref4-3 Suppressors Restore Growth and Soluble Phenylpropanoid Accumulation to Varying Extents.

(A) Sinpoylmalate content from 3-week-old whole rosettes. Data represent the mean ± sd (n = 5). Asterisk indicates P < 0.05 when compared with ref4-3 by Dunnet’s test (Supplemental File 1).

(B) Representative photograph of 3-week-old wild0type (Col-0), ref4-3, med5a, med5b, med5ab, and ref4-3 suppressors (rfs) grown under a long-day photoperiod (16 h light/ 8 h dark, 100 μE m⁻² s⁻¹ halogen and fluorescent light).

Intragenic Missense Mutations in Conserved Residues Suppress the ref4-3 Phenotype

The ref4-3 mutation is semidominant, and plants carrying loss-of-function mutations in MED5b do not display any of its phenotypes, suggesting that intragenic hypomorphic med5b mutations might be present within our pool of suppressors. With this in mind, we first screened a subset of the suppressors for mutations in MED5b using Sanger sequencing. This approach identified five suppressors with mutations in MED5b (Table 1; Supplemental Table 1 and Supplemental Figure 1). Two of these suppressors, rfs8-3 and rfs8-6, carried the same mutation but were derived from the same M1 pool and may result from the same mutagenic event.

Table 1. ref4-3 Suppressor Mutations Identified by Sequencing.

Gene	Suppressor	Allele	Sequencing Method	Codon/Nucleotide Substitutiona	Amino Acid Substitution
MED2	rfs7-4	med2-3	Next Gen	TGG:TGA	W48stop
MED5b	rfs31-1	med5-2	Next Gen	gAA:aAA	Splice site acceptor
	rfs33-2	med5b-3	Sanger	CGC:CAC	R387H
	rfs8-3	med5b-4	Sanger	CTT:TTT	L807F
	rfs8-6	med5b-4	Sanger	CTT:TTT	L807F
	rfs3-3	med5b-5	Sanger	CCT:TCT	P917S
	rfs8-2	med5b-6	Sanger	CCT:CTT	P919L
	rfs6-2	med5b-7	Next Gen	GAA:AAA	E1312K
MED16	rfs7-3	med16-2	Next Gen	TGG:TGA	W898stop
	rfs8-5	med16-3	Next Gen	TGG:TGA	W898stop
MED23	rfs2-2	med23-1	Next Gen	TGG:TAG	W129stop
	rfs9-4	med23-2	Next Gen	TGG:TGA	W259stop
	rfs28-2	med23-3	Next Gen	AGg:AAg	Splice site donor
N.D.	rfs2-1	N.D.	Next Gen	N.D.	N.D.

N.D., not determined. Supplemental Table 1 provides a complete list of alleles identified or used in this study. A schematic diagram of the alleles is provided in Supplemental Figure 1.

^aLowercase letters indicate nucleotides located in introns.

To identify mutations within the rest of the suppressors, we sequenced their genomes using Illumina next-generation sequencing technology and identified all positions that deviate from the reference genotype. Each of the suppressors had ∼2300 mutations compared with the TAIR10 assembly of the Arabidopsis reference genome sequence. To narrow down the list of candidate mutations in each line, we looked for homozygous, likely EMS-induced GC:AT transitions, that generated non-silent mutations in protein coding sequences. On average, ∼280 candidate mutations met these criteria in each line. Examination of the whole-genome sequence data identified two additional suppressors (rfs6-2 and rfs31-1) carrying intragenic mutations in MED5b (Table 1).

Of the six unique intragenic suppressor mutations identified, five introduce missense mutations and one (rfs31-1/med5b-2) is likely to produce a null allele by disrupting splicing of the second intron. Like the med5b-1 T-DNA insertion mutant, med5b-2 ref4-3 is indistinguishable from the wild type in both its growth and sinapoylmalate content, indicating that med5b-2 is probably a null mutant. The other intragenic suppressors restored sinapoylmalate content and growth to varying extents (Figure 1). An alignment of 47 predicted MED5 orthologs from 17 plant species (Supplemental Data Set 1) showed that although only 20% of the residues in Arabidopsis MED5b are conserved across most of the sequences we examined, four of the five missense mutations affect residues that are highly conserved (Figure 2).

Figure 2.

The Intragenic ref4-3 Suppressor Mutations Affect Highly Conserved Residues.

Sequence logos generated from an alignment of 47 MED5 sequences from 17 plant species. The residues mutated in each of the suppressors are indicated below the logos by their positions in the Arabidopsis MED5b sequence and the corresponding amino acid substitution. ref4-3 (G383) and rfs33-2/med5b-3 (R387H) (A), rfs3-3/med5b-5 (P917S) and rfs8-2/med5b-6 (P919L) with WW domain binding motif pattern indicated above the sequence logo (B), rfs8-3/rfs8-6/medb5-4 (L807F) (C), and rfs6-2/med5b-7 (E1312K) (D).

A previously reported intragenic suppressor, rfs8-2/med5b-6, has a proline-to-leucine substitution at residue 919 that restores both growth and sinapoylmalate content to wild-type levels (Stout et al., 2008). Here, we found that one of the newly sequenced suppressors, rfs3-3/med5b-5, has a mutation two residues away from the med5b-6 mutation that introduces a proline-to-serine substitution (P917S). Linear motif prediction (Dinkel et al., 2016) suggests that these prolines may constitute part of a Class IV WW domain binding motif (DWPSPA), a ligand for WW domains that selectively bind phosphorylated p(S/T)P sites (Figure 2B; Sudol and Hunter, 2000). Intriguingly, the P(S/T)P motif was also recently shown to be enriched among phosphosites targeted by human CDK8, a subunit of the Mediator complex (Poss et al., 2016).

The other two intragenic suppressors that affect conserved residues, rfs8-3/rfs8-6/med5b-4 and rfs6-2/med5b-7, change the polarity or charge of amino acids and may disrupt the structure and/or function of ref4-3 by altering intra- or intermolecular interactions (Table 1, Figures 2C and 2D). The last intragenic suppressor, med5b-3/rfs33-2, introduces a mutation four residues away from the ref4-3 mutation (Figure 2A; Bonawitz et al., 2012). As we previously reported, the med5b-3 mutation converts arginine 387 to a histidine, which appears to be reversion to the ancestral sequence at that position (Bonawitz et al., 2012). The med5b-3 mutation is not a null mutation, as it restores growth but maintains the repression of sinapoylmalate accumulation seen in ref4-3, an observation that again demonstrates that these phenotypes can be uncoupled (Figures 1A and 1B).

Disruption of MED2, MED16, or MED23 Suppresses ref4-3

Seven suppressors did not have mutations in MED5b, so we next tested the hypothesis that disruption of other MED subunits might also suppress the function of ref4-3. All but one of the remaining suppressors had a mutation in a gene encoding a subunit of the Mediator complex. These included a mutation in MED2 (rfs7-4), two genetically independent suppressors with identical mutations in MED16 (rfs7-3 and rfs8-5), and three independent suppressors with mutations in MED23 (rfs2-2, rfs9-4, and rfs28-2) (Table 1). The med16 and med23 mutants all restored sinapoylmalate content and growth in the ref4-3 background. The med2 mutant, similar to the med5b-3 intragenic suppressor, restored growth but did not restore sinapoylmalate content (Figure 1). One suppressor, rfs2-1, did not have a mutation in any of the MED genes and its identity remains to be determined.

To test whether the mutations identified in MED2, MED16, and MED23 were responsible for the suppression of ref4-3, we performed complementation tests between the putatively allelic suppressors or additional alleles (Figure 3). The two suppressors with alleles of MED16 failed to complement each other, indicating that they are defective at the same locus. The three suppressors with MED23 alleles also failed to complement each other, demonstrating that they are also allelic. In the case of MED2, we only isolated one allele, so we performed the complementation test using a med2-1 T-DNA insertion mutant. We constructed a med2-1 ref4-3 double mutant, which was not dwarf, initially suggesting that the rfs7-4 suppressor isolated from our screen is likely to be a bona fide med2 mutant. Furthermore, the F1 progeny resulting from test crosses between rfs7-4 and med2-1 ref4-3 double mutants were similar to the parents in size and sinapoylmalate content, thus confirming this initial conclusion. Hereafter, rfs7-4 will be referred to as the med2-3 allele of MED2 (Figure 3A). Finally, we obtained T-DNA mutants of MED16 (sfr6-2; Knight et al., 2009) and MED23 (med23-4) and used them to generate double mutants with ref4-3, as we had done with MED2, and found that these loss-of-function mutations suppressed ref4-3 phenotypes. Based on these observations, we conclude that the mutations that we identified in MED2, MED16, and MED23 in the ref4-3 suppressors are responsible for the suppression of ref4-3.

Figure 3.

ref4-3 Suppressors with Mutations in the Same MED Genes Are Allelic.

(A) Representative photographs and sinapoylmalate content of 3-week-old wild type (Col-0), ref4-3, and F1 from crossed ref4-3 suppressors. Sinapoylmalate quantification represent the mean ± sd (n = 5). * and ‡ indicate P < 0.05 (Dunnet’s test) when compared with ref4-3 and Col-0, respectively.

(B) Sinapoylmalate content quantified from suppressors, med T-DNA lines, and med ref4-3 double mutants. Quantification and statistics are the same as in (A), except n = 4.

The T-DNA mutants also provided the opportunity to examine the impact of the MED mutations on phenylpropanoid metabolism and growth in an otherwise wild-type background. The med2-1 and med16 (sfr6-2) mutants accumulated wild-type levels of sinapoylmalate, whereas the med23-4 mutant had increased sinapoylmalate content compared with the wild type, similar to the med5ab double mutant (Figure 3B). All of the individual mutants were of normal height (Figure 4). RNA-seq analysis of the mutants (described in more detail below) confirmed that in the med5ab, med16 (sfr6-2), and med23-4 mutants, full-length transcripts of the respective genes were essentially abolished (Supplemental Figure 2). In the med2-1 mutant, the T-DNA insertion caused transcription of MED2 to be initiated in an intron in the 5′ untranslated region (Supplemental Figure 2B). There are three AUG codons where translation could start in this newly transcribed region, but all of them are out of frame with the wild-type protein. Thus, med2-1 transcripts likely do not produce functional proteins.

Figure 4.

Height of med Mutants and ref4-3 Suppressors at Maturity.

Representative photographs were taken 9.5 weeks after planting.

Disruption of MED25 Does Not Suppress ref4-3

In addition to MED2, MED5, MED16, and MED23, the Arabidopsis Mediator tail module is also thought to include the MED3, MED15, and MED25 subunits. MED25 is one of the most well studied Arabidopsis MED subunits and has been found to be required for the regulation of numerous processes (Kazan, 2017). Recently, disruption of MED25 was found to suppress the short hypocotyl phenotype of the high sugar response8-1 mutant, indicating a role for MED25 in suppression of cell wall elongation in response to cell wall defects (Seguela-Arnaud et al., 2015). We therefore decided to test whether MED25 is also required for suppression of growth and phenylpropanoid accumulation in the ref4-3 mutant. We obtained two previously reported T-DNA mutants of MED25/PFT, pft1-2 and pft1-3, and used them to generate pft1-2 ref4-3 and pft1-3 ref4-3 double mutants (Kidd et al., 2009). The pft1-2 ref4-3 and pft1-3 ref4-3 mutants were indistinguishable from ref4-3 in both their growth and sinapoylmalate content (Figure 5), indicating that disruption of MED25 is not sufficient to suppress ref4-3. These data demonstrate that ref4-3 does not require the presence of all tail subunits to inhibit growth or phenylpropanoid metabolism.

Figure 5.

Mutation of MED25 Does Not Restore Growth or Sinapoylmalate Accumulation in the ref4-3 Background.

(A) Representative photographs of 3-week-old wild type (Col-0), ref4-3, pft1-2, pft1-3, and double mutants.

(B) Sinapoylmalate content quantified from 3-week-old rosettes. Data represent the mean ± sd (n = 4). Asterisk indicates P < 0.05 (Dunnet’s test) when compared with ref4-3.

Total Lignin Content Is Increased in Some of the Suppressors Compared with ref4-3

As we have previously shown, ref4-3 plants are dwarf and deposit less lignin than wild-type plants, with no change in ratio of the different lignin monomers (Stout et al., 2008). The dwarfism of ref4-3 could be the direct result of reduced lignin content or modified gene expression resulting from perturbations in phenylpropanoid metabolism or other processes (Bonawitz et al., 2014). The med2-1 ref4-3, med2-3 ref4-3, and med5b-3 ref4-3 double mutants restored growth but not sinapoylmalate content relative to ref4-3 (Figures 3B and 4); thus, we asked whether some suppressors might restore growth without restoring lignin accumulation. Lignin content in ref4-3, the suppressors, and the corresponding T-DNA lines were determined by both the acetyl bromide and thioglycolic acid methods (Figure 6) (Campbell and Ellis, 1992; Chang et al., 2008). As previously observed, the ref4-3 mutant showed a 30 to 35% reduction in total lignin content compared with the wild type. In most of the suppressors, the average total lignin content was between that of wild type and ref4-3, but could not be statistically distinguished from either. The same was true for the med2-1 and med16 (sfr6-2) single mutants. One exception to this was med23-1 ref4-3, which by both methods, completely restored lignin content to wild-type levels. Several of the suppressors were significantly different from either the wild type (e.g., med5b-3 ref4-3) or ref4-3 (e.g., med16-3 ref4-3, med23-2 ref4-3, and med23-3 ref4-3), but only when measured by one method or the other. Taken together, our data suggest that lignin content is partially restored in some of the suppressors.

Figure 6.

Total Lignin Content Is Restored to Varying Degrees in the ref4-3 Suppressors

Quantification of total lignin content by the thioglycolic acid method (A) and the acetyl bromide method (B). Data represent the mean ± sd (n = 4). * and ‡ indicate P < 0.05 (Dunnet’s test) when compared with ref4-3 and Col-0, respectively.

All of the ref4-3 suppressors achieved a final height comparable to that of wild-type plants (Figure 4). Several failed to stand upright at maturity, suggesting that they had weaker inflorescences. This was particularly noticeable for med2-3 ref4-3, and to a lesser extent med16-3 ref4-3 and the med2-1 single mutant (Figure 7). Given the variable suppression of lignin content in these mutants (Figure 6), it is not clear that this phenotype is related to lignin content.

Figure 7.

Growth Phenotypes of the med Mutants and ref4-3 Suppressors.

Photos were taken when the inflorescences had reached ∼15 to 25 cm above the soil. Bar = 5 cm.

Transcriptomic Analysis Shows That ref4-3 and med5ab Do Not Have Strictly Opposing Effects on Gene Expression

Given that ref4-3 and its suppressors carry mutations in different subunits of Mediator, an important coregulator of transcription, we sought to determine how the mutations affect gene expression, individually and in concert, and whether we could identify genes that might contribute to the phenotypic differences of the mutants. To do this, we performed whole-transcriptome sequencing (RNA-seq) of ref4-3, the suppressors, the corresponding med T-DNA lines, and wild-type plants. To minimize the possibility that other EMS-induced mutations carried by the suppressors might affect the transcription of some genes, we performed the experiment using the F1 plants generated for complementation testing. These F1 individuals are trans-heterozygous at the suppressor locus, homozygous at ref4-3, and presumably heterozygous for all other EMS-induced changes found in the original suppressors. For simplicity, we refer to these genotypes as med2 ref4-3 (med2-1/med2-3 ref4-3/ref4-3), med16 ref4-3 (med16-2/med16-3 ref4-3/ref4-3), and med23 ref4-3 (med23-1/med23-2 ref4-3/ref4-3).

To determine how similar the gene expression profiles of the mutants were to each other, we used multidimensional scaling to visualize the Euclidian distance between the gene expression profiles of individual samples (Figure 8). In addition to demonstrating that the biological replicates of each genotype were well clustered, the multidimensional scaling indicated that ref4-3 and the med16 (sfr6-2) single mutant had very distinct expression profiles from the rest of the genotypes and from each other (Figure 8A). It also showed that the expression profile of med23 ref4-3 was more similar to that of the wild type than those of the other suppressors, suggesting that med23 is a stronger suppressor of ref4-3 than the other mutants. This conclusion is consistent with the observation that the med23 ref4-3 mutant lines had the highest average sinapoylmalate content of the suppressors (Figures 1A, 3A, and 3B). Pearson’s correlation coefficients calculated from the expression profiles showed that of most of the genotypes were positively but weakly correlated with one another (Figure 8B). The noticeable exceptions to this were ref4-3 and med5ab, which had a correlation coefficient of −0.14, consistent with their opposing effects on the phenylpropanoid pathway (Bonawitz et al., 2012).

Figure 8.

Correlation of Gene Expression Profiles for ref4-3, the Suppressors, and the Corresponding T-DNA Mutants.

(A) Multidimensional scaling plot of individual RNA-seq samples. Values correspond to distances between expression profiles based on the root mean square of the largest absolute log fold changes (leading log fold change) between each pair of samples.

(B) Pearson’s correlation (r) matrix of the gene expression profiles for each genotype. Gene expression profiles include all genes determined to be differentially expressed by the ANOVA-like procedure implemented in EdgeR (false discovery rate <0.05). Clustering was done using Ward’s method.

Given that ref4-3 and med5ab have opposing phenylpropanoid phenotypes (Bonawitz et al., 2012), we determined whether the majority of genes that were differentially expressed in both mutants exhibited the same opposing pattern. In ref4-3, 7074 genes were differentially expressed when compared with the wild type, and in med5ab, 3080 were differentially expressed (adjusted P < 0.05). Of these genes, 1820 (22%) were differentially expressed in both mutants, 58% of which were differentially expressed in opposite directions in the two mutants when compared with the wild type (Figure 9). If we restricted our analysis to the genes whose expression was changed at least 2-fold, 67% were upregulated in ref4-3 and downregulated in med5ab. Gene Ontology (GO) term enrichment analysis of these genes showed enrichment for stress- and hormone-responsive genes. Significantly enriched GO categories (adjusted P ≤ 0.05) included “response to jasmonic acid,” “regulation of systemic acquired resistance,” “response to abscisic acid,” “response to wounding,” “response to chitin,” “defense response,” and “defense response to fungus.” Relatively few genes were downregulated in ref4-3 and upregulated in med5ab (Figure 9). The only GO category that was enriched for this set of genes was “cellular response to iron ion starvation” (5.2E-05), which included three BHLH transcription factors (BHLH038, BHLH039, and BHLH100). There were also many genes that were differentially expressed in the same direction in ref4-3 and med5ab (Figure 9). Genes that were upregulated at least 2-fold in both genotypes were enriched for genes involved in “cell differentiation” (3.8E-03), and genes that were downregulated in both lines were enriched for genes related to the “ethylene activated signaling pathway” (9.2E-03) and “vasculature development” (2.8E-02).

Figure 9.

med5ab and ref4-3 Gene Expression Profiles Show Both Positively and Negatively Correlated Genes.

Scatterplot includes only those genes that are differentially expressed compared with the wild type (P < 0.05) in both ref4-3 and med5ab. Genes with a log₂ fold change ≥ 1 are highlighted in red.

med2, med16, and med23 Alleviate the Widespread Transcriptional Reprogramming of ref4-3

If the changes in gene expression observed in ref4-3 were responsible for the mutant’s altered growth and metabolite accumulation, we expected that the Mediator mutants we identified might suppress ref4-3 by inhibiting these changes. Indeed, most of the 1268 genes that were differentially expressed by more than 2-fold in ref4-3 were no longer differentially expressed or exhibited a less than a 2-fold change in the suppressors (Figure 10). Of the 512 genes that were downregulated in ref4-3, 113 of them were restored to wild-type levels, or were upregulated compared with the wild type, in all three suppressors (Figure 10B). Similarly, 341 of the 756 genes that exhibited increased expression in ref4-3 were either decreased in expression compared with the wild type, or no longer differentially expressed, in all of the suppressors (Figure 10C). These results are consistent with a model in which the suppressors restore growth and phenylpropanoid content by inhibiting the effects of ref4-3 on transcription.

Figure 10.

Most of the Genes That Are Differentially Expressed in ref4-3 Are Restored to Wild-Type Levels in the Suppressors.

(A) Total number of genes that are differentially expressed at least 2-fold when compared with the wild type (P < 0.05) in each line. Direction indicates increased or decreased expression compared with the wild type. Black bars indicate the number of genes that are differentially expressed in ref4-3, or both ref4-3 and the suppressors; dark-gray bars indicate the number of genes that are differentially expressed in opposite directions in ref4-3 and the suppressors; light-gray bars indicate the number of genes that are differentially expressed in the suppressors but not differentially expressed in ref4-3. DEG, differentially expressed genes.

(B) Number of genes that are upregulated in ref4-3 and downregulated, or not differentially expressed, in each of the suppressors.

(C) Number of genes that are downregulated in ref4-3 and upregulated, or not differentially expressed, in each of the suppressors.

To determine how MED2, MED5a/b, MED16, and MED23 contribute to global gene expression in the absence of ref4-3, we examined the effects of the individual med mutants on gene expression (Figure 10A). In the med23-4 mutant, few genes were differentially expressed. In med2-1, as in med5ab, significantly more genes were decreased in expression than were increased in expression, consistent with their roles as coactivators (Figure 10A) (Myers et al., 1999; Ito et al., 2002; van de Peppel et al., 2005). The med16 (sfr6-2) mutant was unique in that a similar number of genes were upregulated as were downregulated. This is consistent with results from yeast, which suggest that MED16 contributes to both gene activation and repression (Covitz et al., 1994; Jiang and Stillman, 1995; Chen et al., 1993). Surprisingly, a large number of differentially expressed genes shared by med16 and ref4-3 were not similarly misexpressed in med16 ref4-3 (Figure 10A).

The ref4-3 Suppressors Restore the Expression of Growth-Promoting Genes and Mitigate the Expression of Stress-Responsive Genes in the ref4-3 Background

To determine whether the suppressors affect gene expression in similar or different ways in the ref4-3 background, we performed hierarchical clustering on all of genes that were differentially expressed at least 2-fold in ref4-3 or the suppressors when compared with the wild type (Figure 11). From this analysis, several patterns of expression emerged. Gene clusters 2, 8, and 9 contained genes that were differentially expressed in ref4-3 and that were restored to a similar extent in all of the suppressors. Clusters 8 and 9 included genes that were downregulated in ref4-3 and were enriched for growth regulators and genes involved in cell expansion. The genes in cluster 2 were upregulated in ref4-3 and enriched for genes involved in response to abiotic stress and, in particular, water stress. Cluster 11 contained genes that were similarly downregulated in ref4-3 and all of the suppressors and thus is unlikely to contribute to their phenotypic differences.

Figure 11.

Hierarchical Clustering Analysis of Genes That Are Differentially Expressed in ref4-3 or the Suppressors.

Heat map displays all genes that are differentially expressed in at least one of the lines, at least 2-fold, when compared with the wild type (P < 0.05). Statistically significant (false discovery rate < 0.05) GO terms for each cluster are listed at the right. The dashed line indicates where the tree was cut to determine cluster membership.

As previously mentioned, med23 appears to be the strongest suppressor of ref4-3. Consistent with this, clusters 1, 3, 4, and 5 contained genes that were upregulated in ref4-3 and more strongly suppressed in med23 ref4-3 than in the other suppressors. Cluster 4 is particularly notable in that it contained only 45 genes and was enriched in phenylpropanoid-related genes, including KFB39, UGT72E2, BGLU22, and a peroxidase superfamily gene (AT2G38390). The increased expression of KFB39 and UGT72E2 in the ref4-3 mutant is consistent with the reduced level of phenylpropanoids in the mutant, as the products of these genes negatively impact the accumulation of lignin, sinapoylmalate, and other soluble phenylpropanoids (Lanot et al., 2006; Zhang et al., 2015).

ref4-3 Promotes the Expression of Negative Regulators of Phenylpropanoid Biosynthesis

To assess the effect of ref4-3 and the suppressors on the transcriptional regulation of phenylpropanoid metabolism, we next examined the expression of phenylpropanoid biosynthetic genes as well as genes encoding transcriptional or posttranslational regulators of phenylpropanoid and secondary cell wall biosynthesis (Figure 12). As previously mentioned, a number of phenylpropanoid biosynthetic genes are upregulated in med5ab and downregulated in ref4-3 (Bonawitz et al., 2012). In this study, we found that genes of the early steps of the phenylpropanoid pathway, such as PAL, C4H, 4CL, HCT, and C3′H, had moderately reduced expression in ref4-3 and some of the suppressors. These genes had moderately increased expression in med5ab double mutants and the med16 (sfr6-2), and to lesser extent, med23-4 single mutants (Figure 12A). As was apparent from the GO term enrichment analysis, many flavonoid biosynthetic genes and transcriptional regulators of flavonoid biosynthesis were upregulated in ref4-3 (Supplemental Figure 3). This observation would appear to be at odds with the fact that ref4-3 has reduced flavonoid accumulation (Bonawitz et al., 2012) and suggests that in the mutant, phenylpropanoid biosynthesis may be limited prior to the branching of the phenylpropanoid and flavonoid pathways. The shikimate pathway supplies phenylalanine to the phenylpropanoid pathway; however, most of the shikimate pathway genes were not differentially expressed in ref4-3 (Supplemental Figure 4). These observations suggest that, in ref4-3, inhibition of phenylpropanoid biosynthesis likely occurs downstream of phenylalanine biosynthesis, but prior to the conversion of p-coumaroyl-CoA to p-coumaroyl-shikimate.

Figure 12.

Differential Expression of Phenylpropanoid-Related Genes in ref4-3, the Suppressors, and the MED T-DNA Mutants Compared with the Wild Type.

Values indicated the log₂ fold change in expression when compared with the wild type. All values are significantly different from the wild type (adjusted P ≤ 0.05). ANOVA P value indicates the significance of a gene as determined by one-way ANOVA across all genotypes, including the wild type.

Many of the genes encoding transcriptional and posttranslational regulators of phenylpropanoid and secondary cell wall biosynthesis (Zhong et al., 2010; Taylor-Teeples et al., 2015) were either not expressed in the tissue we examined (e.g., SND1, NST1, NST2, MYB63, and MYB61) or were not differentially expressed (e.g., MYB58, MYB85, and MYB46; Supplemental Data Set 2). Of the genes that we queried, only MYB4, KFB01, and KFB39 were differentially expressed at least 2-fold in one or more of the mutants (Figure 12B). MYB4 negatively regulates the expression of C4H (Jin et al., 2000) and in the ref4-3 mutant, expression of MYB4 was increased and that of C4H was decreased (Figure 12B; Supplemental Figure 5). In med5ab plants, the opposite was true. Expression of MYB4 was somewhat elevated in med2-1 and the ref4-3 suppressor mutants and was negatively correlated with the expression of C4H (r = −0.9) (Supplemental Figure 5). These data suggest that the suppressors may partially mitigate the ref4-3 metabolic phenotype by impairing the upregulation of MYB4 in the ref4-3 background.

KFB01, KFB20, KFB39, and KFB50 encode F-box proteins that regulate the ubiquitination and subsequent degradation of the four PAL isozymes (Zhang et al., 2013, 2015). KFB01, KFB39, and KFB50 were upregulated in ref4-3, with KFB39 showing the greatest increase (13-fold) in expression (Figure 12B). In the suppressors, KFB39 upregulation was diminished to an extent consistent with the restoration of sinapoylmalate accumulation in each line (r = −0.72). KFB39 was also downregulated in med5ab more than 4-fold, which may contribute to the increased sinapoylmalate content in that line. Together, these data suggest that activation of MYB4 and proteasome-mediated turnover of PAL contribute to the reduced phenylpropanoid content of ref4-3.

DISCUSSION

The Mediator complex is required for the regulation of numerous processes in plants, including development, flowering, phenylpropanoid metabolism, tolerance to freezing, pathogen defense, and responses to light and nutrient availability (reviewed in Samanta and Thakur, 2015; Yang et al., 2016; Dolan and Chapple, 2017). To date, the majority of studies in Arabidopsis have focused on the role of individual MED subunits in these processes. In this study, we took a forward genetic approach and, in doing so, found that additional MED subunits are involved in the transcriptional regulation of phenylpropanoid metabolism. Specifically, we found that MED2, MED16, and MED23 are required for the repression of growth and/or phenylpropanoid metabolism in ref4-3, a semidominant MED5b mutant. With this approach we were also able to identify specific residues in MED5b, that when mutated, suppress the effects of the ref4-3 mutation on phenylpropanoid metabolism. These findings provide strong genetic data for the interdependency of multiple Mediator subunits in the establishment of the ref4-3 phenotype. Furthermore, although the screen is not completely saturated, the retrieval of such a high proportion of Mediator subunit suppressors suggests that the architecture of downstream processes involved in ref4-3 dwarfing and metabolic perturbations may be highly branched or redundant.

MED2, MED16, and MED23 Are Required for the Function of ref4-3

The architecture of the plant Mediator complex has not been extensively studied but our finding that mutations in MED2, MED16, and MED23 can suppress the effects of a semidominant MED5b mutation provides genetic support for a model of plant Mediator that is similar to that of yeast and humans. Structural and functional analyses of human Mediator have shown that the homologs of the MED2, MED5, MED16, and MED23 subunits physically and functionally interact with one another in the tail module (Ito et al., 2002; Tsai et al., 2014), as do yeast MED2, MED5, and MED16 (Béve et al., 2005; Robinson et al., 2015). These subunits have also been shown to share some functions in Arabidopsis, wherein MED2, MED16, and MED14 are required for cold-regulated gene expression, and MED5a/b, MED16, and MED14 are required for dark-induced gene expression (Hemsley et al., 2014). Our finding that many stress response genes are upregulated in ref4-3 and downregulated in med5ab suggests that MED5 and MED16 might also share some functions in the transcriptional regulation of stress response pathways (Boyce et al., 2003; Wathugala et al., 2012). Aside from med5a/b, only med23-4 had an effect on phenylpropanoid accumulation, independent of ref4-3 (Figure 3B). Thus, the requirement of MED2 and MED16 for the repression of phenylpropanoid biosynthesis may be unique to ref4-3. Alternatively, if ref4-3 emulates the structural or biochemical modifications through which MED5 normally downregulates phenylpropanoid biosynthesis in response to internal and external cues, then MED2 and MED16 may be relevant to the wild type as well.

Analyses of knockout mutants made in yeast and mouse cell lines showed that MED2, MED5, MED16, and MED23 depend on one another to various degrees for their association with Mediator. Electron microscopy of yeast and human Mediator suggests that MED5 is the most distal of the tail subunits, although the precise locations of MED23 and MED25 within the tail module have not been determined (Tsai et al., 2014; Robinson et al., 2015). In mouse embryonic fibroblasts, deletion of MED5 resulted in the loss of MED16, MED23, and a large fraction of CDK8 from the complex (Ito et al., 2002). Similarly, deletion of MED23 resulted in reduced association of MED16 and MED5 (Stevens et al., 2002). In yeast, knocking out Med16 causes dissociation of the rest of the tail from the complex (Zhang et al., 2004; Béve et al., 2005; Tsai et al., 2014). These observations suggest that the MED16 and MED23 mutations we identified might suppress the function of ref4-3 by displacing it from the complex. This model is consistent with the observation that med23-4 has increased sinapoylmalate content, similar to med5ab (Figure 3B). Although we did not observe an increase in sinapoylmalate content in med16 (sfr6-2) (Figure 3B), several of the phenylpropanoid biosynthetic genes are upregulated and KFB20, KFB39, and KFB50 are all downregulated in that line as in med5ab (Figure 12). Why med16 does not show an increase in sinapoylmalate content is not clear, but misregulation of other genes in the med16 mutant might mask the effects of changes in KFB-mediated PAL degradation. In contrast to MED23 and MED16, MED2 does not appear to be as important for the integrity of the tail module. Deletion of yeast Med2 results in loss of Med3 (Myers et al., 1999; Béve et al., 2005), but not Med5 or Med16 (Béve et al., 2005), suggesting that med2 might suppress ref4-3 by a different mechanism such as by disrupting interactions with transcription factors or RNA Polymerase II or by changing the conformation or flexibility of the complex, either inherently or in response to interactions with these other partners. Identification of the transcription factors that interact with these subunits will enable a better understanding of how their perturbation might affect phenylpropanoid metabolism, growth, or other processes. To date, MED2 has been shown to interact with GeBP, and MED16 with FIT and WRKY33 (Zhang et al., 2014; Wang et al., 2015; Shaikhali et al., 2016). So far, no transcription factors have been shown to target MED5a/b or MED23.

Intragenic Suppressors Identify Functionally Important Residues in MED5b

How the ref4-3 mutation, which causes a single glycine-to-serine substitution, alters the function of Mediator is unknown. The mutation could alter interactions between MED5b and other MED subunits, transcription factors, transcription coregulators, or components of the general transcription machinery. Alternatively, ref4-3 might disrupt the overall conformation of the complex or interfere with its conformational flexibility, an essential feature of Mediator (Taatjes et al., 2004; Tóth-Petróczy et al., 2008). We identified five intragenic suppressors of ref4-3 that introduce missense mutations in conserved residues of MED5b (Figure 2). Some of these may be null mutations that disrupt the folding or activity of MED5b; however, several observations suggest that some may have relatively specific effects on the function of the protein. First, several of the mutations are very close to one another. The med5b-3 mutation is four residues away from the ref4-3 mutation and is not null because although it restores growth, med5b-3 still represses sinapoylmalate accumulation (rfs33-2 in Figures 1A and 1B). Two additional intragenic mutations (rfs3-3/med5b-5 and rfs8-2/med5b-6) affect proline residues that are separated by a single serine residue. The clustering of these mutations suggests that these regions might be particularly important for the function of MED5b. Second, both of these regions contain a PXP motif, one of which is predicted to form part of a Class IV WW domain binding motif (Figures 2A and 2B; Sudol and Hunter, 2000). Proline-rich motifs impart unique secondary structures and are a common ligand of intracellular signaling molecules (Zarrinpar et al., 2003). Disruption of these motifs in the suppressors might alter the local structure of the protein or affect its interactions with other proteins.

ref4-3 Appears to Limit Phenylpropanoid Biosynthesis by Mimicking a Common Regulatory Mechanism Employed by MED5a/b

We have previously shown that MED5a/b limits phenylpropanoid biosynthesis in response to genetic perturbations in the phenylpropanoid and glucosinolate pathways, but the mechanisms underlying these phenomena are not entirely understood (Bonawitz et al., 2014; Anderson et al., 2015; Kim et al., 2015). In the absence of MED5a/b, Arabidopsis hyperaccumulates soluble phenylpropanoids, possibly due to loss of this repressive pathway (Bonawitz et al., 2012). Conversely, ref4-3 constitutively represses phenylpropanoid accumulation (Stout et al., 2008; Bonawitz et al., 2012). The opposing phenotypes of ref4-3 and med5ab, and their opposing effects on expression of the phenylpropanoid biosynthetic genes (Figure 12A), suggest that ref4-3 acts as a hypermorph, at least with regard to phenylpropanoid metabolism. When we compared the effects of the ref4-3 and med5ab mutations on gene expression more broadly, we found that, while a large number of genes are differentially expressed in opposite directions in the two mutants, there are also a substantial number of genes that are differentially expressed in the same direction (Figure 9). Although we cannot distinguish which of these changes is the direct result of the mutations in MED5 and which are indirect effects, on the whole, these data suggest that ref4-3 acts as a hypermorph in the context of some genes, such as the phenylpropanoid genes, and as a hypomorph in the context of others.

The observation that many more genes are downregulated than upregulated in the med5ab mutant suggests that MED5a/b functions primarily as a transcription coactivator (Figure 10A). Therefore, the best candidates for direct targets of MED5a/b are genes that are downregulated in med5ab, such as MYB4 and KFB39. This supports a model in which MED5a/b limits phenylpropanoid biosynthesis by activating transcription of genes encoding negative regulators of the pathway rather than, or in addition to, direct repression of biosynthetic gene transcription. In transgenic lines overexpressing KFB39 to a similar extent as in ref4-3, PAL activity was reduced by ∼40% and sinapate ester accumulation was reduced by more than half (Zhang et al., 2015), indicating that the transcriptional changes observed in ref4-3 are likely to have a significant impact on phenylpropanoid accumulation. Given that all carbon that enters the phenylpropanoid pathway must first be channeled through PAL, turnover of the PAL proteins would quickly limit flux into the pathway in response to changes in metabolism or the environment. Indeed, others have shown that the KFB genes are downregulated in the presence of high levels of sucrose and following treatment with UV light (Zhang et al., 2013, 2015). A model involving downregulation of PAL would also explain the reduction of flavonoids in ref4-3 despite upregulation of flavonoid biosynthetic genes.

Multiple Pathways Affecting Plant Growth Are Misregulated in ref4-3

The ref4-3 mutant is dwarf and has reduced lignin content, but the relationship between dwarfism and the reduction in lignin is unclear (Bonawitz and Chapple, 2013). Although growth was restored in all of the suppressors (Figure 4), in most of the lines lignin content appeared to be only partially restored (Figure 6). Like many low lignin mutants, ref4-3 exhibits collapsed xylem cells due to reduced lignification of the vasculature, which has been proposed to lead to reduced water conductivity and consequently impaired growth (Stout et al., 2008; Bonawitz and Chapple, 2013). Consistent with this, genes involved in response to abiotic stress and in particular, water stress, were upregulated in ref4-3 (Figure 11, Cluster 2), as has been observed in the lignin biosynthetic mutant ref8-1 (Bonawitz et al., 2014). In the suppressors, these genes exhibit wild-type levels of expression, suggesting that lignin accumulation might be sufficiently restored in those lines to prevent vascular collapse and alleviate water stress. Nevertheless, several of the suppressors still have low sinapoylmalate content (Figure 3), demonstrating that, in some respects, phenylpropanoid biosynthesis and growth can be genetically disentangled. The mutants identified here will serve as useful genetic material for determining whether different signaling pathways underlie these distinct phenotypes or if they are the result of varying thresholds for the impact of differential gene expression on phenylpropanoid biosynthesis and growth.

The dwarfism of ref4-3 may be the result of activation of one or more growth-repressing pathways. Several low-lignin mutants have been shown to accumulate increased levels of salicylic acid (SA), and blocking SA accumulation restored growth in Medicago sativa and Arabidopsis HCT RNAi lines (Gallego-Giraldo et al., 2011; Lee et al., 2011); however, this does not appear to be a universal response of low-lignin mutants, as the ref8-1 mutant does not accumulate increased levels of SA, and blocking SA biosynthesis has no effect on growth in that line (Bonawitz et al., 2014). Another possibility is that the reduced lignin content in ref4-3 triggers activation of a cell wall integrity pathway, which in turn results in reduced growth, similar to what has been observed for some cellulose-deficient mutants (Turner, 2007; Bonawitz and Chapple, 2013). Recently, med16 was found to suppress the growth defects of the cellulose-deficient mutant cobra, demonstrating that the Mediator tail module is important for growth inhibition in response to cell wall defects (Sorek et al., 2015); MED5 could play a similar role as a link between cell wall integrity and phenylpropanoid biosynthesis.

METHODS

Plant Materials and Growth

Arabidopsis thaliana (ecotype Columbia-0) was grown in Redi-earth Plug and Seedling Mix (Sun Gro Horticulture) at a temperature of 23°C, under a 16-h-light/8-h-dark photoperiod with a light intensity of 100 μE m⁻² s⁻¹ supplied by a combination of halogen and fluorescent bulbs. All plants were grown nine per 4 × 4-inch pot unless otherwise noted. After planting, seeds were held for 2 d at 4°C before transferring to the growth chamber.

The semidominant ref4-3 mutant was isolated from a forward genetic screen for plants deficient in sinapoylmalate accumulation and has been described previously (Ruegger and Chapple, 2001; Stout et al., 2008; Bonawitz et al., 2012). Genotyping of the ref4-3 (Stout et al., 2008), ref8-1, and ref8-2 alleles was performed as previously described (Bonawitz et al., 2014).

The rfs mutants were isolated from a forward genetic screen in which ∼75,000 ref4-3 seeds were mutagenized with EMS and the suppressors selected from the M2 generation based on their restored growth or sinapoylmalate accumulation, as previously described (Stout et al., 2008). Salk insertion lines were obtained from the ABRC (Ohio State University) unless otherwise noted. The insertion lines used in this study are as follows: med5b-1/ref4-6 (SALK_ 037472), med5a-1/rfr1-3 (SALK_011621) (Bonawitz et al., 2012), med2-1 (SALK_023845C) (Hemsley et al., 2014), sfr6-2 (SALK_048091) (Knight et al., 2009), med23-4 (SALK_074015C), pft1-3 (SALK_059316) (Kidd et al., 2009), med3-1 (SALK_015589C), and med3-2 (SALK_012449). The med2-1, med3-1, med3-2, med23-4, and pft1-3 mutants were generously provided by Tesfaye Mengiste (Department of Botany and Plant Pathology, Purdue University). The med16-1/sfr6-2 mutant was kindly provided by Zhonglin Mou (Department of Microbiology and Cell Science, University of Florida). Salk lines were genotyped and confirmed homozygous, as described at http://signal.salk.edu/tdnaprimers.2.html. The primers used for genotyping are listed in Supplemental Table 2.

HPLC Quantification of Soluble Metabolites

Sinapate esters and flavonols were extracted from 3-week-old whole rosettes at 65°C for 1 h at a concentration of 100 mg of tissue (fresh weight) per mL 50% methanol. Samples were centrifuged in a microcentrifuge at 16,000g for 5 min to sediment plant material. A 10-μL sample of the supernatant was run on a Shim-pack XR-ODS column (Shimadzu; column dimensions 3.0 mm × 75 mm, bead size 2.2 µm) and eluted with a gradient of 2 to 25% acetonitrile in 0.1% formic acid (v/v) over 30 min, at a flow rate of 0.9 mL min⁻¹. Sinapic acid was used as a standard for the quantification of sinapoylmalate.

Lignin Quantification

Inflorescence tissue was harvested for total lignin quantification after senescence was complete, with each biological replicate consisting of six to nine plants pooled from a single pot. Stem tissue was cleaned of leaves and siliques, chopped roughly, and then ground to a powder using a Retsch MM400 ball mill grinder. Alcohol-insoluble cell wall residue was prepared by incubating the ground stem tissue with 1 volume of 0.1 M sodium phosphate buffer, pH 7.2, at 50°C for 30 min, followed by washing with 70% ethanol at 70°C until all visible traces of chlorophyll were gone. The residue was then rinsed with acetone and left to dry overnight. Total lignin was quantified from the cell wall residue by both the acetyl bromide and thiolglycolic acid methods as previously described (Campbell and Ellis, 1992; Chang et al., 2008).

DNA Extraction and Whole-Genome Sequencing

For Sanger sequencing, genomic DNA was extracted from leaves and amplified with BigDye terminator v3.1 (Life Technologies), using the primers listed in Supplemental Table 3. PCR products were then sent to the Purdue Genomics Core Facility for sequencing (Purdue University).

For whole-genome sequencing, genomic DNA was isolated from young leaf tissue by grinding with ceramic mortar and pestle under liquid nitrogen, followed by extraction of nucleic acids using a CTAB/sarkosyl and chloroform:isoamyl alcohol procedure published by Diversity Arrays Technology (http://www.diversityarrays.com/sites/default/files/pub/DArT_DNA_isolation.pdf). Approximately 200 mg of ground tissue was transferred to a 1.5-mL microfuge tube to which was added 600 μL of fresh buffer solution. Fresh buffer solution was made by mixing 1 volume extraction buffer (0.35 M sorbitol, 0.1 M Tris HCl, pH 8.0, 5 mM EDTA, 0.5% [w/v] sodium metabisulfite, and 2% [w/v] polyvinylpyrrolidone-40), 1 volume lysis buffer stock (0.2 M Tris HCl, pH 8.0, 0.05 M EDTA, 2 M NaCl, and 2% CTAB), and 0.4 volumes of 5% (w/v) sarkosyl. Samples were vortexed vigorously in buffer preheated to 65°C and then incubated at 65°C for 1 h, with inversion every 20 min. Samples were then allowed to cool to room temperature and 600 μL of chloroform:isoamyl alcohol (24:1) was added before shaking for 30 min. After shaking, samples were centrifuged at 16,000g in a microcentrifuge for 10 min. The aqueous phase was transferred to a new microfuge tube and the DNA was precipitated by adding the same volume of −20°C isopropanol (350 µL) and gently inverting six times. The DNA was pelleted by centrifugation at 16,000g for 10 min. DNA pellets were washed with 500 μL of 70% ethanol. After washing, ethanol was removed and pellets were allowed to air dry for an hour. Dried pellets were dissolved in 50 μL of Tris-EDTA, pH 8.0.

Whole-genome sequencing was performed by the Purdue Genomics Core Facility (Purdue University) on an Illumina HiSeq 2500 instrument. Genomic DNA libraries for all samples were bar-coded and loaded onto one lane for 100-bp, paired-end sequencing. Sequences were aligned to the TAIR10 reference genome sequence using Bowtie2 (Langmead and Salzberg, 2012). Single nucleotide polymorphisms (SNPs) were called using samtools followed by bcftools (Li, 2011). SNPs were annotated using SnpEff and then filtered for homozygous, G-to-A or C-to-T SNPs of “Moderate” or “High” impact using SnpSift (Cingolani et al., 2012a, 2012b).

RNA Extraction and Whole-Transcriptome Sequencing

Samples were collected for whole-transcriptome sequencing (RNA-seq) 20 d after planting, 6.5 h after subjective dawn. For each biological replicate, five whole rosettes were pooled from five different pots with randomized locations and immediately flash frozen in liquid nitrogen. Samples were then stored at −80°C until RNA extraction. For RNA extraction, the pooled rosettes were ground to a powder under liquid nitrogen using a chilled mortar and pestle. Approximately 80 mg of ground tissue was then transferred to an Eppendorf tube for RNA extraction using the RNEasy Plant Mini kit from Qiagen. Total RNA was submitted to the Purdue Genomics Core Facility (Purdue University) for purification of poly(A)+ RNA, library construction, and sequencing. All samples were dual bar-coded, pooled, and loaded onto four sequencing lanes. Paired-end, 100-bp sequencing was performed by an Illumina HiSeq 2500 machine run in “rapid” mode. Read mapping was performed by the Purdue Genomics Core using the TAIR10 genome build and the TopHat alignment program (Trapnell et al., 2009). For Supplemental Figure 2, FPKM values were determined using Cufflinks v2.2.1 with default parameters. The RNA-seq data discussed in this publication have been deposited in NCBI’s Gene Expression Omnibus (Edgar et al., 2002) under GEO Series accession number GSE95574.

Statistical Analysis of RNA-Seq Data

For each sample sent for RNA sequencing, digital gene expression (counts) for every exon was determined using the HTSeq-count program with the intersection “nonempty” option (Anders et al., 2015). Counts were summarized by gene ID. The edgeR program was used for differential gene expression analysis (Robinson et al., 2010). The edgeR analysis began with a count table comprising 33,602 genes. Genes expressed at low levels were then filtered out by removing any genes for which there was not at least one count per million reads in at least four of the samples. This resulted in a list of 18,842 expressed genes (Supplemental Data Set 3). Next, the “ANOVA-like” procedure in edgeR was used to identify genes that were differentially expressed between any of the genotypes (false discovery rate < 0.05). This resulted in a list of 14,688 differentially expressed genes (Supplemental Data Set 3A). To determine which of these genes were differentially expressed in each of the mutants compared with the wild type, the exact test for the negative binomial distribution (Robinson and Smyth, 2008) was used followed by a Bonferroni correction to maintain a gene-wise error rate of 0.05 (Supplemental Data Set 3B). GO analysis was performed using the DAVID Bioinformatics Resource v6.8 (Huang et al., 2009a, 2009b). All genes that were expressed in our experiment were used as the set of background genes to test for ontology enrichment.

Protein Sequence Analysis

Forty-seven MED5 coding sequences (CDSs) from 17 species were collected from NCBI and Phytozome (Supplemental Table 4). MED5 CDSs were translated to protein sequences using the SEAVIEW (v4.5.4) program (Gouy et al., 2010), and alignment of the protein sequences was performed using the MUSCLE alignment algorithm (Edgar, 2004) (Supplemental Data Set 1). Sequence logos indicating the residue frequency at each position were created from this alignment using the WebLogo program (http://weblogo.berkeley.edu) (Crooks et al., 2004). The Arabidopsis MED5b CDS sequences in the NCBI, TAIR, and Phytozome databases did not agree with our transcriptome sequencing; therefore, we generated a new MED5b CDS sequence from our data, which we have used in our analyses and deposited in GenBank under accession number KY593335.

Accession Numbers

Sequence data can be found under the following Arabidopsis Genome Initiative accession numbers: MED2 (At1g11760), MED3 (At3g09180), MED5a/RFR1 (At3g23590), MED5b/REF4 (At2g48110), MED16/SFR6 (At4g04920), MED23 (At1g23230), and MED25/PFT1 (At1g25540). RNA-seq data have been deposited in NCBI’s Gene Expression Omnibus (Edgar et al., 2002) under GEO Series accession number GSE95574. The new MED5b CDS sequence generated from our RNA-seq data has been deposited in GenBank under accession number KY593335.

Supplemental Data

• Supplemental Figure 1. Schematic Diagram of Alleles Identified or Used in This Study.

• Supplemental Figure 2. Full-Length Transcripts Are Substantially Reduced in the med T-DNA Mutants.

• Supplemental Figure 3. Differential Expression of Flavonoid Biosynthetic Genes and Transcription Factors in ref4-3, the Suppressors, and the MED T-DNA Mutants Compared with the Wild Type.

• Supplemental Figure 4. Differential Expression of Shikimate Biosynthesis Genes in ref4-3, the Suppressors, and the MED T-DNA Mutants Compared with the Wild Type.

• Supplemental Figure 5. The Expression of MYB4 and C4H Are Negatively Correlated.

• Supplemental Table 1. Mutant Alleles Used in This Study.

• Supplemental Table 2. Primers Used for Genotyping SALK Lines.

• Supplemental Table 3. Primers Used for Sanger Sequencing of MED5B.

• Supplemental Table 4. Accessions Used for MED5 Multiple Sequence Alignment.

• Supplemental Data Set 1. Multiple Protein Sequence Alignment of MED5 Homologs.

• Supplemental Data Set 2. Differential Expression of Phenylpropanoid Related Genes.

• Supplemental Data Set 3. Results of Differential Gene Expression Analysis.

• Supplemental File 1. ANOVA Tables.