The Proteasome Stress Regulon Is Controlled by a Pair of NAC Transcription Factors in Arabidopsis

Proteotoxic stress in Arabidopsis is attenuated by a pair of NAC transcription factors that upregulate the synthesis of the 26S proteasome and other factors that promote protein homeostasis.

Abstract

Proteotoxic stress, which is generated by the accumulation of unfolded or aberrant proteins due to environmental or cellular perturbations, can be mitigated by several mechanisms, including activation of the unfolded protein response and coordinated increases in protein chaperones and activities that direct proteolysis, such as the 26S proteasome. Using RNA-seq analyses combined with chemical inhibitors or mutants that induce proteotoxic stress by impairing 26S proteasome capacity, we defined the transcriptional network that responds to this stress in Arabidopsis thaliana. This network includes genes encoding core and assembly factors needed to build the complete 26S particle, alternative proteasome capping factors, enzymes involved in protein ubiquitylation/deubiquitylation and cellular detoxification, protein chaperones, autophagy components, and various transcriptional regulators. Many loci in this proteasome-stress regulon contain a consensus cis-element upstream of the transcription start site, which was previously identified as a binding site for the NAM/ATAF1/CUC2 78 (NAC78) transcription factor. Double mutants disrupting NAC78 and its closest relative NAC53 are compromised in the activation of this regulon and notably are strongly hypersensitive to the proteasome inhibitors MG132 and bortezomib. Given that NAC53 and NAC78 homo- and heterodimerize, we propose that they work as a pair in activating the expression of numerous factors that help plants survive proteotoxic stress and thus play a central regulatory role in maintaining protein homeostasis.

INTRODUCTION

All cellular organisms require mechanisms to dampen the accumulation of aberrant proteins caused by transcription/translation errors, misfolding, cleavage, chemical modification, and environmental conditions such as heat that perturb tertiary/quaternary structures. If allowed to hyperaccumulate, this detritus can negatively impact a host of intracellular activities through an imbalance in protein homeostasis and excessive protein aggregation (Morimoto, 2008; Hipp et al., 2014). The resulting proteotoxic stress has strong physiological consequences, including genome instability, arrest of the cell cycle, inhibition of translation through a reduction in ribosomes, lost membrane integrity, inhibition of metabolism, and accelerated senescence, as well as amino acid starvation if protein recycling becomes severely compromised. In fact, the accumulation of misfolded, aggregation prone proteins is the hallmark of a broad range of human diseases (Cuanalo-Contreras et al., 2013).

To alleviate this stress, plants, fungi, and animals elicit a number of cytoprotective responses, including expression of protein chaperones and activation of the unfolded protein response designed to promote refolding and sequester protein aggregates, maintenance of chromatin integrity through SUMOylation, and upregulation of several proteolytic pathways that remove these aberrant polypeptides before they become cytotoxic (Hetz, 2012; Howell, 2013; Kim et al., 2013; Amm et al., 2014; Seifert et al., 2015). Excessive protein aggregation and the accumulation of defective protein complexes (e.g., ribosomes and proteasomes) and organelles (e.g., chloroplasts and mitochondria) often engage autophagy (Li and Vierstra, 2012; Khaminets et al., 2016; Marshall et al., 2015). These damaged structures are encapsulated into cytoplasmic vesicles and delivered to the vacuole/lysosome for breakdown, in many cases using ubiquitylation as a signal.

Arguably, the most important protease upregulated during proteotoxic stress is the 26S proteasome (Amm et al., 2014; Hanssum et al., 2014; Livnat-Levanon et al., 2014). This 2.5-MD, ATP-dependent proteolytic machine works in tandem with ubiquitin (Ub) to direct the selective breakdown of aberrant polypeptides and normal short-lived proteins in the cytoplasm and nucleus. Targets are first covalently modified with multiple Ubs; the resulting ubiquitylated species dock with the 26S proteasome, which degrades the modified protein concomitant with release of the Ub moieties for reuse.

The 26S proteasome consists of two subcomplexes: the 20S core protease (CP) and the 19S regulatory particle (RP) (Finley, 2009; Bhattacharyya et al., 2014). The CP consists of four stacked heteroheptameric rings of distinct α- and β-subunits in an α_1-7/β_1-7/β_1-7/α_1-7 configuration, which houses in a central chamber the proteolytic active sites provided by the PBA (β1), PBB (β2), and PBE (β5) subunits. The RP has 18 or more subunits; it caps one or both ends of the CP barrel and contains receptors for ubiquitylated targets and activities that remove the Ub moieties and unfold and translocate the target polypeptides into the CP lumen before breakdown. In addition, a host of accessory factors associate substoichiometrically, including dedicated chaperones that promote the sequential assembly of the CP and RP complexes and final construction of the 26S particle, alternative capping factors (PA200 and CDC48), shuttle proteins that help deliver ubiquitylated substrates, and ubiquitin-protein ligases (or E3s) and deubiquitylating enzymes (Finley, 2009; Hanssum et al., 2014).

Given its role(s) in maintaining protein homeostasis, the levels of the 26S proteasome are highly regulated to meet demand, which is achieved by the coordinated expression of all core subunits and most, if not all, accessory factors. In yeast (Saccharomyces cerevisiae), this regulon is driven by the C2H2-type zinc-finger transcription factor Rpn4, which binds to the PACE (proteasome associated control element) cis-element found upstream of most proteasome subunit genes (Mannhaupt et al., 1999; Xie and Varshavsky, 2001; Shirozu et al., 2015). Rpn4 responds to proteotoxic stress by itself being a target of proteasomal breakdown. When proteolytic demand is low, Rpn4 is rapidly degraded (t_1/2 ∼2.5 min), thus attenuating expression of 26S proteasome genes (Xie and Varshavsky, 2001; Dohmen et al., 2007). However, as 26S proteasome capacity is challenged by proteotoxic stress, Rpn4 is stabilized, thus allowing its levels to rise and upregulate particle synthesis.

Although orthologs of Rpn4 are not obvious outside of yeasts, a similar proteasome stress regulon (PSR) exists in many other eukaryotes, including mammalian cells, Drosophila melanogaster, and Arabidopsis thaliana (Meiners et al., 2003; Yang et al., 2004; Lundgren et al., 2005; Kurepa et al., 2008; Book et al., 2009). In mice and humans, the Nrf1 (Nuclear Factor Erythroid-derived 2-Related Factor 1) transcription factor, unrelated to Rpn4, was recently shown to be a central effector (Radhakrishnan et al., 2010, 2014; Sha and Goldberg, 2014). Nrf1 is bound to the endoplasmic reticulum (ER) and is constitutively degraded by the ER-associated protein degradation pathway, and like Δrpn4 yeast cells, nrf1-null cells are more sensitive to proteasome inhibitors. Upon proteotoxic stress, Nrf1 is proteolytically released from the ER and is now free to enter the nucleus and activate the PSR.

How the PSR is controlled by proteotoxic stress remains unclear in plants. One candidate regulator in Arabidopsis is NAC78, a member of the NO APICAL MERISTEM/ARABIDOPSIS TRANSCRIPTION ACTIVATION FACTOR1/CUP-SHAPED COTYLEDONS2 (NAC) family of transcriptional regulators. NAC78 was first implicated by overexpression studies showing that it positively regulates the expression of core 26S proteasome subunit genes and that its putative DNA binding site is present within many, but not all, associated promoters (Yabuta et al., 2011; Nguyen et al., 2013). Whereas NAC78-OX plants are smaller than the wild type, nac78 mutant plants are larger, which is consistent with a reported role for 26S proteasome levels in controlling cell size (Kurepa et al., 2009; Sonoda et al., 2009).

To more broadly define the plant PSR, we combined RNA-seq analysis with both chemical inhibitors and proteasome mutants that induce proteotoxic stress. Gene coexpression together with protein interactome network analyses revealed a complex network of PSR genes/proteins enriched in protein chaperones, autophagy components, and detoxification enzymes that presumably help plants cope with proteotoxic stress, in addition to those encoding the 26S proteasome and its assembly/capping components. Network and promoter-interaction studies then identified the NAC78 paralog NAC53 as a key effector that, together with NAC78, likely functions as homo- and heterodimers. Importantly, seedlings simultaneously lacking NAC53 and NAC78 poorly activate the PSR during proteotoxic stress, and their growth is strongly hypersensitive to proteasome inhibitors. As such, we propose that these two NAC proteins are central regulators of an extended PSR in Arabidopsis by providing sufficient 26S proteasomes and other protein homeostatic factors to mitigate proteotoxic stress.

RESULTS

26S Proteasome Subunits Are Upregulated during Proteotoxic Stress

To better understand how 26S proteasome gene expression is upregulated by proteotoxic stress, we examined the transcript levels for representative Arabidopsis subunits either upon short- (3 h) or long-term (24 h) exposure of seedlings to the proteasome inhibitor MG132 (Yang et al., 2004), or in mutant backgrounds that compromise 26S proteasome assembly (rpn10-1 and rpn12a-1) and elicit seedling phenotypes consistent with impaired capacity (Smalle et al., 2002, 2003). The mutant alleles were fortuitously generated by exon-trap mutagenesis; they are viable but have stunted growth and show pleiotropic defects in hormone signaling. The rpn12a-1 allele dampens expression of the full-length RPN12a transcript (see Figure 2A) but is relatively mild phenotypically (Smalle et al., 2002). The rpn10-1 allele expresses a truncation of RPN10, the main Ub receptor in the 26S complex, and has much stronger developmental consequences (Smalle et al., 2003). The translated polypeptide includes the N-terminal von Willebrand Factor-A domain that links RPN10 to the rest of the RP but is missing the C-terminal region containing the three Ub-interacting motifs that bind Ub, the autophagy adaptor ATG8, and cargo receptors bearing Ub-like domains, respectively (Farmer et al., 2010; Fatimababy et al., 2010; Marshall et al., 2015).

Figure 2.

Expression of Arabidopsis Proteasome Genes Is Upregulated in Response to Proteasome Stress.

(A) Expression of 26S proteasome subunit genes following proteasome inhibition with MG132 or in rpn10-1 or rpn12a-1 mutant backgrounds that compromise assembly. Total RNA from 5-d-old seedlings, either untreated or treated for 3 and 24 h with 100 µM MG132, was subjected to RT-qPCR. The expression values were calculated using the ACT2 transcript as a reference and normalized to those from untreated wild-type seedlings. Each bar represents the average of at least three biological replicates (±sd).

(B) Effects of MG132 on the expression of proteasome promoter:GUS transgenes. Transgenic wild-type seedlings expressing the fusions were grown for 3 d, treated for 1 d with 100 µM MG132, and then incubated overnight with the X-Gluc substrate.

(C) Quantitative measure of proteasome promoter:GUS expression following MG132 treatment. Ten-day-old seedlings were incubated overnight with or without 100 µM MG132 and homogenized, and the GUS activity in the resulting cell extracts was assayed using the MUG substrate. Each bar represents the analysis of at least 30 independent T1 lines, each assayed in triplicate (±sd). The data in (B) and (C) for the RPT2a and RPT2b promoter:GUS fusions were reported previously and are included here for comparison (Lee et al., 2011).

Previous studies demonstrated that both long-term exposure to MG132 and the rpn10-1 mutation elevate the steady state levels of Ub conjugates and increase the abundance of several core subunits of the 26S proteasome (Smalle et al., 2003; Yang et al., 2004; Kurepa et al., 2008). We extended these results here by immunoblotting crude extracts from MG132-treated wild-type and untreated homozygous rpn10-1 and rpn12a-1 seedlings with antibodies against Ub, the CP subunit PBA1(β1), the RP subunits RPN1, RPT2, and RPN12a, and the CP regulator PA200 (known as Blm10 in yeast) (Figures 1A and 1B). In particular, accumulation of the unprocessed form of PBA1 increased strongly upon MG132 treatment, consistent with the need for active proteasomes to generate the mature, truncated β1 polypeptide (Finley, 2009; Book et al., 2010). The rpn12a-1 allele had only a marginal effect on the abundance of Ub conjugates and the PBA1, RPN1, and RPT2 subunits, but its action was obvious based on its strong effect on PA200 levels (Figure 1A). Why the effects on CP/RP subunit levels were mild for the rpn12a-1 allele was unclear, but it could reflect the relatively modest phenotype of the mutant (Smalle et al., 2002) and/or the possibility that genetically compromised 26S proteasomes generated by the rpn12a-1 allele more effectively induce autophagic turnover of the complex as opposed to the rpn10-1 mutation, which blocks such turnover (Marshall et al., 2015).

Figure 1.

Both the Proteasome Inhibitor MG132 and Proteasome Mutants Increase the Accumulation of Proteasome Components and Ub Conjugates in Arabidopsis.

Wild-type seedlings treated with 100 µM MG132 or untreated rpn10-1 and rpn12a-1 seedlings were grown for 5 d. Total extracts were probed by immunoblotting with the indicated antibodies, using anti-histone H3 antibodies to verify near equal protein loading.

(A) Levels of individual subunits of the proteasome. The open and closed arrowheads identify the unprocessed and processed forms of the β1 subunit PBA1.

(B) Levels of Ub conjugates. Closed arrowheads locate free Ub and poly-Ub chains assembled with varying numbers of Ub monomers. Bracket indicates high molecular mass Ub conjugates.

When mRNA abundance was then examined by RT-qPCR analysis of seedlings, levels of the CP subunit PBA1 and the RP subunits RPN5a, RPN10, and RPN12a were found to rise rapidly upon short treatments with MG132 and were constitutively upregulated in the rpn10-1 and rpn12a-1 backgrounds (Figure 2A). (Throughout this study, we used transcripts from the ACTIN2 [ACT2] and/or TYPE-2A SERINE/THREONINE PROTEIN PHOSPHATASE [PP2A] genes as controls given their relative immunity to proteotoxic stress [Supplemental Figure 1].) As with previous studies (Gallois et al., 2009; Lee et al., 2011), this increased transcript abundance could also be demonstrated with transgenic plants expressing the GUS reporter under the control of various 26S proteasome gene promoters. Upregulation upon MG132 treatment was visualized colorimetrically by staining the seedlings with X-Gluc and quantitatively by 4-methylumbelliferyl-β-d-glucuronide (MUG)-based fluorescence activity assays of crude seedling extracts (Figures 2B and 2C). Interestingly, comparisons of several gene pairs that encode individual proteasome subunits revealed that often only one responds to MG132, implying that the many pairs have subfunctionalized with one locus mainly responsible for increasing subunit mRNA levels during proteotoxic stress (e.g., RPN3a, RPT1a, RPT2a, and RPT4b; Figures 2B and 2C).

RNA-Seq Analysis of the Proteasome Stress-Induced Regulon

To more fully identify the suite of genes that are upregulated when proteasome capacity is compromised, we performed RNA-seq analysis with our cohort of wild-type seedlings treated with MG132 (3 and 24 h) and untreated rpn10-1 and rpn12a-1 seedlings. These transcriptome studies identified a large collection of mRNAs whose abundance was significantly affected as compared with untreated wild-type seedlings based on negative binomial normalization using edgeR differential expression analysis (P value < 0.01, false discovery rate [FDR] < 0.05; Supplemental Data Sets 1 and 2). By merging the data sets, we identified 119 genes (including RPN10 and RPN12a given their upregulation in two of the three conditions) that were coordinately upregulated under all three conditions (3-h treatment with MG132 and in the rpn10-1 and rpn12a-1 backgrounds), which we designated as members of the PSR (Figure 3A; Supplemental Data Set 3). RNA-seq analysis after a 24-h exposure to MG132 identified an additional set of 865 upregulated genes; these loci were not included in the final PSR as their long-term induction might be more indirectly related to the prolonged effects of the inhibitor and the downstream responses to severe proteotoxic stress (Supplemental Data Set 4).

Figure 3.

Characterization of the PSR in Arabidopsis.

(A) Venn diagrams of transcripts that were differentially expressed significantly (P value < 0.01, FDR < 0.05) under three proteasome stress conditions: the wild type after a 3-h treatment with 100 µM MG132, and the rpn10-1 and rpn12a-1 genetic backgrounds. The shared 119 upregulated (including RPN10 and RPN12a) and 33 downregulated genes determined by edgeR analysis comprise the PSR.

(B) A heat map of the PSR for the three treatments ranked by the fold change in expression obtained with the MG132-treated wild-type seedlings. The brackets indicate genes with more or less than a 2-fold increase (log₂(fold change) = 1) in expression compared with the wild type. Categories of enriched gene classes are indicated. TFs, transcription factors. See Supplemental Data Set 3 for the full list.

(C) Expression change heat maps of 26S proteasome subunit and assembly factor genes after proteasome stress compared with untreated wild type. EST values obtained from The Arabidopsis Information Resource (TAIR v10) are listed.

Gene Ontology analyses via DAVID (database for annotation, visualization, and integrated discovery; https://david.ncifcrf.gov) of the 119 PSR upregulated genes detected a significant functional enrichment for the 26S proteasome, Ub-related events, and general stress response processes and identified several transcription factors that might transcriptionally activate the PSR. This enrichment for 26S proteasome genes was especially strong in the more robustly affected group with a ≥2-fold increase (44% of total; Figure 3B). Of the 53 genes encoding core 26S proteasome subunits (Book et al., 2010; Russell et al., 2013), 48 (91%) were significantly upregulated (Figures 3B and 3C). We note that the rank order of the upregulated and downregulated genes in the MG132 and rpn10-1/rpn12a-1 data sets based on their strength of change differed markedly (Figure 3B); this deviation could reflect two distinct subtypes of proteotoxic stress, one elicited rapidly and strongly by the inhibitor and the other being a more subtle chronic stress elicited by the mutations. We also detected 33 genes whose expression was downregulated by all three conditions (Figures 3A and 3B; Supplemental Data Set 3); there was no significant functional enrichment in this group, thus leaving their collective action(s) unclear.

From analysis of genes known or predicted to be associated with the Arabidopsis 26S proteasome (Finley, 2009; Book et al., 2010), a strikingly coordinated upregulation was evident for most loci. This was apparent for both core CP and RP subunits as well as acknowledged accessory factors (PA200 and ECM29) and chaperones (HSM3, NAS6, PBAC1, PBAC2, and UMP1a) that are presumably needed during proteotoxic stress to assist in particle regulation/assembly (Figure 3C). The proteasome binding protein PROTEASOME REGULATOR1, recently identified as a positive regulator of auxin signaling through activation of the 26S proteasome (Yang et al., 2016), was also upregulated, as was the possible plant-specific chaperone PAP1. The only exception for subunits encoded by a single gene was the locus encoding the Ub receptor RPN13, whose expression was relatively immune to MG132 and was unchanged by the mutations. In agreement with the RT-qPCR and proteasome promoter:GUS reporter studies above, we found that multiple gene pairs encoding individual subunits had at least one locus that was PSR regulated with the second locus sometimes nonresponsive. Examples include the PAA2, PBB2, RPT1a, RPT6b, RPN3a, and RPN9b loci, which were highly sensitive to proteotoxic stress, whereas their paralogs PAA1, PBB1, RPT1b, RPT6a, RPN3, and RPN9a were comparatively silent (Figure 3C).

When the analysis was extended to nonproteasome transcripts, additional loci potentially relevant to the PSR were detected (Supplemental Data Set 3). Included were mRNAs encoding members of the NAC, WRKY, and heat shock protein transcription factor (HSF) families that often have prominent roles in various plant stress responses (Rushton et al., 2010; Nakashima et al., 2012; Scharf et al., 2012), other Ub/proteasome pathway components including Ub, several Ub E3 ligases (AIP2, At1g71020, At4g27050, and At5g27920), deubiquitylating enzymes (UBP6, UBP7, and UCH3), extraproteasomal Ub binding proteins (DSK2a and DDI1), and components of the autophagy system (Supplemental Figure 2). Notable autophagy factors included several members of the ATG8 family required for autophagic vesicle dynamics and cargo delivery and the autophagy receptor NBR1, which is devoted to the recruitment of ubiquitylated cargo (Li and Vierstra, 2012; Zhou et al., 2013). Strong transcriptional upregulation of NBR1, especially by MG132, implied a role for autophagy in removing Ub conjugates that become stabilized when proteasome capacity is limited.

The MG132-Induced PSR Network

To better visualize the MG132-induced PSR, we entered the 336 MG132-induced genes that were significantly upregulated ≥2-fold after a 3-h treatment (Supplemental Data Set 1) into the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) coexpression and interactome database (Jensen et al., 2009). Upon visualization in Cytoscape v3.0 (Shannon et al., 2003), an extensive collage of networks involving 275 output loci emerged that clustered into three main groups (Figure 4A). As expected, one tight cluster included loci encoding 26S proteasome subunits and accessory factors/proteasome chaperones. A second cluster encompassed a large collection of general heat shock protein (HSP) chaperones (several HSP70 isoforms, HSA32, HSP81, and HSP101, and their dedicated transcriptional regulator HSF8A) that presumably direct protein refolding and/or minimize aggregation of insoluble proteins awaiting proteasomal/autophagic turnover (Lin et al., 2014). A third more diffuse cluster was enriched in the stress-related NAC, WRKY, and HSF transcription factors that might help activate the PSR and transferase activities (e.g., GSTs and O-methyltransferases) that could mitigate proteotoxic stress (Livnat-Levanon et al., 2014). The third cluster also contained several UDP-glycosyl transferases and the main mitochondrial alternative oxidase, AOX1a, which were shown previously to be upregulated upon mitochondrial perturbation (De Clercq et al., 2013). Many of these loci were also seen in clusters generated with the 24-h MG132 treatment data set, especially for the 26S proteasome and related factors (Supplemental Figure 3). Additions to the 24-h MG132 network included a pronounced autophagy cluster and a large cluster enriched in protein kinases that could extend the PSR signaling network given the roles of some in stress defense and developmental signal transduction cascades (e.g., MEK1, MAPKKK1, RLP24, HLECRK, CRCK1, and CRCK11).

Figure 4.

Description of the MG132 Upregulated Gene Interaction and Coexpression Network.

(A) An interaction and coexpression map of Arabidopsis proteins encoded by MG132 upregulated genes. The connections reflect known protein/protein interactions and coexpression data collected from the STRING database for the available 275 of the total 336 significantly upregulated genes (≥2-fold up compared with untreated wild type) after a 3-h exposure to 100 µM MG132. Classifications of major functional groups in the network are highlighted. Members of the NAC transcription factor family are in red. Portions of the map enclosed by the dashed lines are statistically enriched for genes with the denoted functional categories (DAVID, P value < 0.01). If available, specific gene names were used instead of the TAIR locus identifier. See Supplemental Data Set 1 for the full list.

(B) Members of the network that contain the consensus PRCE or MDM cis-motifs within their promoter regions (motif-containing gene nodes are in red).

(C) Sequence descriptions of the PRCE and MDM cis-motifs as determined by MEME.

(D) Interactome maps focusing on the statistically significant central hubs present in the proteasome, chaperone, and detoxification clusters as determined by MCODE analysis and colored based on the presence of PRCE and/or MDM sequences.

Under the assumption that many of the 336 loci are transcriptionally regulated by a common set of transcription factors, we searched their promoter regions for shared cis-regulatory DNA elements using the motif-based sequence analysis tool MEME (Multiple Expectation Maximization for Motif Elicitation; Bailey et al., 2006). Highly enriched were the Proteasome-Related cis-Element (PRCE) with a TGGGC core sequence, which was previously implicated by Nguyen et al. (2013) in the regulation of proteasome genes, and the mitochondrial dysfunction motif (MDM) associated with oxidative stress responses (De Clercq et al., 2013) (Figures 4B and 4C). Both motifs were prominent in the 26S proteasome gene cluster, with few promoters (14 of 48) predicted to be devoid of either motif (Figure 4D). When placed within the region upstream of the translation start site for genes encoding core 26S proteasome subunits, a wide dispersion of the PRCE and MDM motifs was seen, with some having these elements close to the transcription start site and others placed >1000 bp away (Supplemental Figure 4). Whereas single MDM elements were common, we often detected multiple PRCE-related sequences in tandem, sometimes on opposite DNA strands, suggesting that PRCE is acted upon by DNA binding proteins that work palindromically. We also detected other potentially relevant cis-elements associated with the PSR at lower frequencies (Supplemental Figure 5). Included were the SORLIP2 and T-box motifs that are involved in light-induced developmental responses (Chan et al., 2001; Hudson and Quail, 2003).

The NAC53 and NAC78 Transcription Factors Are Associated with the PSR

To help identify the transcriptional regulators that might coordinate the PSR, we screened a yeast one-hybrid (Y1H) library of ∼1700 known or predicted DNA binding proteins from Arabidopsis by an automated luciferase-based expression system (Gaudinier et al., 2011), using both the PA200 or RPN12a promoter regions as bait (1000- and 253-bp fragments upstream of the translation start codon, respectively). The activity of both promoters is strongly upregulated by MG132 exposure, and accordingly they contain multiple segments related to the consensus PRCE sequence. (It should be noted that the RPN12a promoter is predicted to be short given that only 253 bp separates its translation start site from the coding region immediately upstream.) Among the list of 15 interactors that activated both Y1H assays was NAC53 (At3g10500, also known as NTL4; Supplemental Table 1), a 555-residue NAC protein that we also discovered as a locus significantly upregulated by brief MG132 exposure and positioned by Cytoscape as a central hub within the PSR detoxification cluster (Figure 4A; Supplemental Figure 2). Upon scanning the Arabidopsis genome for related proteins, we identified NAC78 (At5g04410, also known as NTL11/RPX1), a 584-residue NAC protein that is its closest relative (∼72% amino acid sequence identity) among the ∼109 NAC-type transcriptional regulators (Olsen et al., 2005a; Nakashima et al., 2012; Wu et al., 2012). Binding of NAC53 and NAC78 to both promoters was confirmed by directed Y1H assays using 500 bp of the PA200 promoter and 253 bp of the RPN12a promoter as baits.

Intriguingly, several previous studies implicated NAC53 and NAC78 in the PSR and/or in binding PRCE/MDM-type sequences. NAC78 was first linked by overexpression screens that searched for factors that regulate either plant size or seedling responses to light stress and was subsequently shown by transcriptome studies to impact proteasome gene expression (Morishita et al., 2009; Yabuta et al., 2011; Nguyen et al., 2013). Nguyen et al. (2013) further connected NAC78 to the PSR by reporting that it recognizes the same PRCE element identified here as common within the regulon (Figure 3C). A further connection to the PSR was provided by DNA binding studies showing that both NAC53 and NAC78 might also recognize MDM elements found here to be common within PSR loci (De Clercq et al., 2013). Finally, NAC53 was previously linked to heat stress, reactive oxygen species (ROS) production, and senescence, all of which are related to proteotoxic stress through cursory phenotypic analyses of NAC53 mutant and overexpression lines (Lee et al., 2012, 2014).

To provide an evolutionary connection between NAC53 and NAC78, we examined the two loci phylogenetically in the context of the entire NAC family. Bayesian analyses clustered NAC53 and NAC78 together when either the 109 full-length NAC proteins were compared or when just the DNA-binding NAM domains were analyzed, indicating that this pair forms a unique subclade within the NAC family (Figure 5A; Supplemental Figures 6 and 7). In fact, strong amino acid sequence identity was seen throughout the NAM domain (∼93%), implying that they bind the same DNA sequence (Supplemental Figure 8). Close sequence relatives (∼50% sequence identity) were also detected in other plant species, including various monocots and dicots and the seedless plants Selaginella moelendorffii and Physcomitrella patens, but not in the alga Chlamydomonas reinhardtii, implying that the pair has a conserved function in land plants. Relatives of both NAC53 and NAC78 could be found in Arabidopsis lyrata that clustered apart from likely orthologs in other dicots, suggesting that NAC53 and NAC78 originated from a duplication that occurred recently within the Brassicaceae lineage.

Figure 5.

NAC53 and NAC78 Are Closely Related and Physically Interact.

(A) A Bayesian phylogenetic tree of all 109 NAC proteins in Arabidopsis rooted to a P. patens NAC protein (gi_168025227). Nodes highlighted in red indicate the NTL subclass with a predicted transmembrane-spanning motif. The same tree with the names included for each protein is in Supplemental Figure 6, and a text file of the alignment used is presented as Supplemental Data Set 5.

(B) Y2H assays showing that NAC53 and NAC78 homo- and heterodimerize. The full-length proteins were expressed as N-terminal fusions with either the GAL4 activating (AD) or binding (BD) domains. Shown are cells grown on selective medium lacking Leu and Trp, or lacking Leu, Trp, and His, and containing 50 mM 3-AT.

(C) BiFC analysis showing that NAC53 and NAC78 homo- and heterodimerize in planta and partially localize to the nucleus. N. benthamiana leaf epidermal cells were coinfiltrated with plasmids expressing the N- and C-terminal fragments of YFP (nYFP and cYFP, respectively) fused to the N terminus of NAC53 or NAC78. Shown are reconstituted BiFC signals, as detected by confocal fluorescence microscopy of leaf epidermal cells 36 h after infiltration, along with DAPI staining of nuclei (white arrowheads) and a bright-field (BF) image of the cells. Bar = 10 μm.

(D) Y2H assays testing interactions between NAC53 and NAC78 and other NAC proteins within the PSR. The assays were conducted as in (B) with the selective medium lacking Leu, Trp, and His and containing 25 mM 3-AT.

The C-terminal ends of NAC53 and NAC78 are predicted to contain a 24-amino acid membrane-spanning sequence that is also found within 11 other Arabidopsis NAC proteins, which are collectively designated as NTLs (NACs with transmembrane-like region; Kim et al., 2007; Supplemental Figure 6). Like the NAM domain, this putative membrane-spanning region is highly conserved between NAC53 and NAC78 (75% sequence identity), even when aligning their sequences next to that of their closest NTL relative NAC13 (30% full polypeptide sequence identify and 65% NAM domain sequence identity; Supplemental Figure 8). Even within the NTL subfamily, NAC53/NAC78 stand apart phylogenetically, suggesting that they have distinct function(s) (Figure 5A; Supplemental Figures 6 and 7). The role(s) of the membrane-spanning domain is not yet clear; for at least one Arabidopsis NTL, it has been proposed to facilitate binding of the NAC protein to cytosolic membranes, with proteolytic release then permitting nuclear import (Kim et al., 2006; Kim et al., 2007).

NAC53 and NAC78 Regulate PSR Gene Expression

Given the aforementioned connections of NAC53/NAC78 to proteasome gene regulation and their likely ability to bind PRCE-type motifs (Yabuta et al., 2011; De Clercq et al., 2013; Nguyen et al., 2013; this report), we hypothesized that NAC53 and NAC78 work in concert as homo- and heterodimers to upregulate the PSR. As a first validation, we tested whether NAC53 and NAC78 interact, as is common for paralogs within the NAC family (Nakashima et al., 2012; De Clercq et al., 2013). By yeast two-hybrid (Y2H) assays, we detected binding of NAC53 and NAC78 to themselves and to each other, using the Gal4/LacZ pDEST system, thus generating colonies that grew in the absence of His and were resistant to 50 mM 3-amino-1,2,4-triazole (3-AT) (Figure 5B). This assembly of homo- and heterodimers was then confirmed in planta by bimolecular fluorescence complementation (BiFC) with the split YFP system transiently expressed in Nicotiana benthamiana leaf epidermal cells. Only when the NAC53 or NAC78 coding regions were simultaneously expressed as fusions to either the nYFP or cYFP fragments was fluorescence reconstituted (Figure 5C; Supplemental Figures 9A and 9C). Fluorescence was easily detected in the cytoplasm and nucleus, suggesting that the two proteins interact in each compartment.

It had been reported that when a GFP fusion of full-length NAC78 is transiently expressed in onion (Allium cepa) epidermal cells, it mainly localizes to the cytosol but becomes concentrated in the nucleus when expressed without the membrane-spanning region (Morishita et al., 2009). To examine whether proteotoxic stress could recapitulate this redistribution, we attempted to generate transgenic plants that stably express GFP-NAC53 and GFP-NAC78 with or without this transmembrane domain. Unfortunately, no fluorescent signals were detected, in agreement with a previous failure with another NTL, NAC13 (De Clercq et al., 2013). As an alternative, we examined the BiFC signal from the NAC53/NAC78 pair in N. benthamiana cells simultaneously treated with MG132. Unfortunately, under our conditions, MG132 treatment neither impacted the strength nor the distribution of the BiFC signals (Supplemental Figure 9B).

As the next step in connecting at least one of this NAC pair to the PSR, we examined transgenic lines expressing NAC78 from the β-estradiol-inducible XVE promoter, which was provided by the TRANSPLANTA collection (Coego et al., 2014). In our hands, NAC78 mRNA levels rose ∼60-fold after an overnight β-estradiol exposure. As shown in Figure 6, pEST:NAC78 seedlings, but not wild-type seedlings, treated with β-estradiol also strongly increased the mRNA abundance for several representative 26S proteasome subunits, as first observed by Nguyen et al. (2013). Importantly, this upregulation extended to other members of the PSR outside of the core 26S particle, including genes encoding the proteasome accessory factor PA200, the NAS6 assembly chaperone, the UPS component UFD1, and the HSP transcriptional regulator HSF8A. Two other PSR genes encoding the proteasome accessory factor PA200 and the glutathione S-transferase GSTU25 also displayed increases close to significance, consistent with NAC78 more broadly controlling the PSR.

Figure 6.

NAC78 Overexpression Induces the Expression of Some Arabidopsis PSR Genes.

Upregulation of proteasome subunit and other PSR genes in seedlings expressing NAC78 from an estradiol-inducible promoter. Total RNA from 6-d-old wild-type and pEST:NAC78 seedlings incubated for 24 h with or without 10 µM β-estradiol were subjected to RT-qPCR. The expression values were calculated using ACT2 (black) and PP2A (gray) transcripts as references and normalized to those obtained from untreated wild-type seedlings. Bars represent the average of at least three biological replicates (±sd), each measured in triplicate. Asterisks indicate significant differences between pEST:NAC78 and the wild type based on Student’s t test (P value < 0.05). Dashed lines indicate the average values for untreated wild-type seedlings.

Assuming that the NAC53/NAC78 proteins work together, we generated double mutants impacting the pair to observe any adverse effects on PSR regulation. Potentially useful alleles were identified in the SALK collection of T-DNA insertion mutants that either disrupted the coding region downstream from the N-terminal NAM domain (nac53-1), within the NAM domain (nac78-1), just upstream of that for the C-terminal transmembrane sequence (nac53-2), or toward the end of the coding region (nac78-2; Figure 7A). Genomic sequencing mapped the insertion sites to 735 and 372 nucleotides downstream of the translational start site for the nac53-1 and nac78-1 mutations and 87 and 21 nucleotides upstream of the translational stop site for the nac53-2 and nac78-2 mutations, respectively.

Figure 7.

Description of Mutations Impacting NAC53 and NAC78.

(A) Diagrams of the NAC53 and NAC78 transcribed regions. Boxes represent coding regions (colored) and predicted untranslated regions (white). The blue and orange boxes identify the DNA binding NAM domains and the predicted membrane-spanning regions, respectively. Lines represent introns. The positions of the T-DNA insertions are indicated by the red triangles; their exact locations in the amino acid sequences are indicated in Supplemental Figure 8.

(B) RT-PCR analysis of the NAC53 and NAC78 transcripts in the single and double mutants. Total RNA isolated from wild-type or homozygous mutant plants was subjected to RT-PCR using the primer pairs indicated in (A). RT-PCR with primers specific for ACT2 was included to confirm analysis of equal amounts of cDNA.

Transcript analyses by RT-PCR failed to detect the corresponding full-length mRNAs in homozygous nac53-1 and nac78-1 seedlings (Figure 7B). For nac53-1, we also could not detect transcripts downstream of the insertion site, whereas for nac78-1 we could detect a transcript of slightly greater length than expected. Collectively, the RT-PCR data imply that nac53-1 and nac78-1 represent null alleles. For homozygous nac53-2 and nac78-2 seedlings, we detected low levels of near-full length transcripts originating upstream of the insertion sites; these mRNAs included the NAM domain coding sequence. For the nac78-2 mutant, only a small portion of the proposed C-terminal membrane-spanning region domain was eliminated, whereas nearly the entire membrane-spanning region was missing for the nac53-2 mutant (Supplemental Figure 8), suggesting that they represent weaker alleles under the assumption that the full transmembrane domain is not essential (at least for NAC53).

When grown at 24°C under a normal 16-h-light/8-h-dark photoperiod on agar medium or soil, all the single and double mutant combinations germinated well, generated phenotypically normal rosettes, flowered at the same time, and were equally fertile as the wild type, indicating that NAC53 and NAC78 alone or in combination are not necessary for Arabidopsis growth, development, and fecundity under nonstressed conditions. At least under our growth conditions, the mutant seedlings were not larger than the wild type, as had been reported previously for a T-DNA insertion line disrupting NAC78 by itself (Nguyen et al., 2013).

Importantly, RT-qPCR analyses of PSR-related genes revealed that NAC53 and NAC78 combine to activate the PSR. Whereas the nac53-1 mutation had little effect on expression of representative PSR genes after an overnight MG132 treatment, a mild but statistically significant dampening (P value < 0.01) was seen for the nac78-1 line (Figure 8A). Moreover, when the double nac53-1 nac78-1 mutant was tested, a substantially attenuated PSR response was observed. Compromised loci included genes encoding subunits of the 26S particle (PBA1, RPN5a, RPN10, and RPN12a), the accessory factors PA200 and CDC48, the NAS6 chaperone, the UPS factor UFD1, the glutathione S-transferase GSTU25, and the heat shock transcription factor HSF8A (P value < 0.01). For some loci, the nac53-1 nac78-1 combination almost completely ablated the transcriptional upregulation. A smaller but significant drop in PSR gene expression was also seen for the nac53-2 nac78-2 double mutant, consistent with the milder effect of these mutations on the NAC53/NAC78 mRNAs (Figure 8A).

Figure 8.

Loss of NAC53 and NAC78 Compromises Activation of the Proteasome Stress Regulon.

(A) RT-qPCR analysis of representative PSR mRNAs during proteasome stress. Total RNA was extracted from 6-d-old wild-type and nac mutant seedlings after a 24-h incubation with or without 100 µM MG132. Transcript abundance was determined via RT-qPCR using the ACT2 (black) and PP2A (gray) mRNAs as references and normalized to those obtained from untreated wild-type seedlings. Bars represent the average of at least three biological replicates (±sd), each measured in triplicate. Asterisks indicate significant differences between the nac51-1 nac78-1 seedlings and the wild type (+MG132) based on Student’s t test (P value <0.05). Dashed lines indicate the average values for untreated wild-type seedlings.

(B) Increased levels of several 26S proteasome subunits during proteasome stress depend on NAC53 and NAC78. Seedlings were treated with or without MG132 as in (A), and the resulting crude extracts were immunoblotted with the indicated antibodies. Histone H3 was included to confirm near equal protein loading. Open and closed arrowheads identify the unprocessed and processed forms of PBA1, respectively.

The diminished PSR in turn reduced the steady state levels of some of the corresponding proteins. Whereas MG132 treatment increased the seedling abundance of the 26S accessory protein PA200, the RP subunit RPN12a, and the unprocessed form of the CP subunit PBA1 in wild-type seedlings, this increase was substantially weaker in the nac53-1 nac78-1 background (Figure 8B). However, this drop did not appear to translate into a substantial stabilization of Ub conjugates, as the pool of immunodetectable species was only mildly increased in the nac53-1 nac78-1 plants compared with the wild type (Supplemental Figure 10). Consistent with the reduced strength of the nac53-2 and nac78-2 alleles on PSR gene expression, the double mutants had little impact on the RPN12a and PBA1 protein levels and possibly only a mild effect on PA200 (Figure 8B). As the polypeptides derived from the nac53-2 nac78-2 alleles should be missing all or part of the presumed transmembrane domain, respectively, membrane association might not be essential for NAC53/NAC78 function.

Plants Missing NAC53 and NAC78 Are Hypersensitive to Proteotoxic Stress

Phenotypic analysis of homozygous nac53-1 nac78-1 plants in turn revealed that both transcription factors together help Arabidopsis survive proteotoxic stress. Whereas wild-type plants could tolerate long-term exposure to sublethal doses of MG132 beginning at germination, growth of the double mutant plants was substantially arrested (Figures 9A and 9B). In fact, development of nac53-1 nac78-1 seeds stalled soon after germination, and for many, the radicals failed to exit the seed coat after rupture. Quantification of the response also detected a significant growth inhibition (P value < 0.01) for the nac78-1 single mutant and for the weaker nac53-2 nac78-2 double mutant compared with the wild type (Figure 9B).

Figure 9.

Plants Lacking Both NAC53 and NAC78 Are Hypersensitive to Proteasome Inhibitors.

(A) Double homozygous nac53-1 nac78-1 plants are hypersensitive to MG132. Ten-day-old seedlings of the indicated genotypes were germinated and grown on MS medium plus sucrose and containing either DMSO (control) or 30 or 50 µM MG132.

(B) Quantification of fresh weight for seedlings shown in (A).

(C) Growth inhibition of 6-d-old nac53-1 nac78-1 seedlings first germinated on MG132-free medium and then transferred to medium containing 50 µM MG132 2 d after germination.

(D) Double homozygous nac53-1 nac78-1 plants are strongly hypersensitive to bortezomib. Seedlings were germinated and grown for 7 d on various concentrations of bortezomib.

(E) Fresh weight of 7-d-old wild-type, nac53-1 nac78-1, and nac53-2 nac78-2 seedlings grown on 1 µM bortezomib (Btz).

Asterisks in panels (B), (C), and (E) indicate a P value < 0.01 based on one-way ANOVA.

Given that the smaller plants observed for the nac53-1 nac78-1 line upon MG132 exposure might have arisen from defective or delayed germination, we also germinated and grew the seedlings for 6 d in the absence of MG132 and then placed them on inhibitor-containing medium and measured seedling fresh weight 6 d afterwards. Even in this situation, growth of the nac53-1 nac78-1 plants was significantly impaired compared with the wild type (Figure 9C). As MG132 might impact other proteases besides those within the CP, we also tested a second 26S proteasome inhibitor with potentially greater specificity and potency: bortezomib (Kisselev et al., 2012). Here, growth of the nac53-1 nac78-1 plants was severely inhibited at submicromolar concentrations, with a mild but significant effect also seen for the nac53-2 nac78-2 plants (Figures 9D and 9E).

NAC53 and NAC78 Work Together and with Other Members of the NAC Family

While our studies indicated that NAC53 and NAC78 work together, possibly as heterodimers based on our Y2H assays, to activate the PSR, they could also have temporally distinct functions within the response (e.g., early versus late) given that NAC53 but not NAC78 transcripts were increased rapidly by MG132 treatment. To test these possibilities, we compared the time course for PSR activation by MG132 in the single mutant nac53-1 and nac78-1 backgrounds by RT-qPCR analysis of representative PSR loci. As shown in Supplemental Figure 11, the induction time courses were identical (but slightly diminished) for genes encoding the CP and RP subunits, CDC48, the RP chaperone NAS6, and GSTU25, with the nac53-1 nac78-1 double mutant lacking a significant response for all loci tested but GSTU25. Together, the data imply that NAC53 and NAC78 work within the same time frame.

The mild attenuation of the PSR for the nac53-1 nac78-1 plants in response to MG132 as seen with GSTU25 and other loci (e.g., WRKY25) (Figure 8A; Supplemental Figure 11) suggested that other transcription factors also contribute to the PSR. Particularly notable were several other members of the NAC family that were transcriptionally upregulated by the inhibitor, including NAC1, 13, 32, 44, 50, 55, 81/ATAF2, 82, and 87 (Supplemental Figure 2). To test whether they also heterodimerize with NAC53/NAC78, we subjected each to Y2H assays using NAC53 and NAC78 as prey. Interestingly, NAC13 and NAC81, but not the others, generated positive Y2H signals, suggesting that they also contribute to PSR activation in combination with NAC53/NAC78 (Figure 5D). Whereas NAC13 bound to both NAC53 and NAC78, NAC81 only showed an interaction with NAC53 under these conditions. NAC13 is especially intriguing given its ability to recognize MDM cis-elements common within the PSR and its membership in the transmembrane-containing NTL-NAC subfamily like NAC53/NAC78.

DISCUSSION

Given the importance of protein quality control to maintaining healthy cellular functions and the likelihood that plants routinely experience proteotoxic stress (e.g., ER-associated protein degradation; Howell 2013), we considered it likely that plants have strong protective responses. Examples include various environmental insults such as heat, cold, salt, drought, excess light, photooxidative stress, and ROS production that perturb protein folding, carbon and nitrogen starvation that limits synthesis of new polypeptides, pathogen invasion that often coincides with the massive translation of pathogen-associated polypeptides that might express improperly in the host, and exposure to toxins made by the pathogen, such as the proteasome inhibitors epoxomicin and syringolin A that block host protein turnover (Yang et al., 2004; Groll et al., 2008). Our RNA-seq studies on Arabidopsis seedlings induced to experience proteotoxic stress through impairment of 26S proteasome capacity (MG132 inhibition and the rpn10-1 and rpn12a-1 mutants) revealed an intricate network of PSR genes/proteins that presumably represent parts of these protective mechanisms. Many of the associated activities are also important for protein homeostasis in yeast and animals, indicating that they reflect conserved protein homeostatic processes among eukaryotes (Morimoto, 2008; Hipp et al., 2014).

A prominent PSR cluster is the 26S proteasome itself and its associated assembly/regulatory factors that are needed to build a functional holoenzyme, demonstrating that increasing 26S proteasome capacity is one crucial protective mechanism. In fact, given the plethora of core subunits and assembly chaperones required to construct 26S proteasomes, a remarkable coordination of expression has evolved to ensure faithful production of complete particles. As shown by mutational studies affecting individual subunits, substoichiometric accumulation of only one subunit is sufficient to stall construction of the entire complex and induce off-product accumulation of assembly intermediates (Smalle et al., 2002, 2003; Book et al., 2009; Lee et al., 2011). The importance of 26S proteasomes to protein homeostasis is also demonstrated by the fact that nearly all core subunits are encoded by at least one gene that is strongly sensitive to proteotoxic stress. The presence of a nonresponsive paralog likely reflects subfunctionalization of the pair, which might be important for maintaining proteasome levels in specific spatio-temporal contexts.

A second PSR cluster includes a collection of protein chaperones that help protein folding and minimize protein aggregation, along with several HSF transcription factors that promote their expression during stress. Their inclusion was anticipated given the central roles of chaperones in protein quality control and maintaining protein homeostasis (Morimoto, 2008; Hipp et al., 2014). A third PSR cluster is enriched in a collection of enzymes important for cellular detoxification, such as glutathione S-transferases, O-methyltransferases, and AOX1a, the last of which is responsible for alternative respiration in mitochondria. Presumably, these proteins help repair damaged proteins and reduce oxidative stress and ROS. Several autophagy components are also members of the PSR, including ATG8 and the Ub receptor NBR1 (Kraft et al., 2010), thus connecting autophagy to plant protein homeostasis. A role for NBR1 is in agreement with studies on nbr1 mutants; they accumulate substantial amounts of high molecular mass Ub conjugates during heat stress, which possibly represent ubiquitylated protein aggregates awaiting autophagic clearance (Zhou et al., 2013). Interestingly, additional autophagy genes appeared in the network created upon long-term MG132 exposure (Supplemental Figure 3), suggesting that autophagy represents a last line of defense during proteotoxic stress. It should also be emphasized that we discovered a number of genes within the Arabidopsis PSR whose functions remain to be discovered. Once their purposes become clear, it is likely that additional protein homeostatic mechanisms will be revealed.

Combined with previous studies (Yabuta et al., 2011; Nguyen et al., 2013), our network and promoter interaction studies identified NAC53 and NAC78 as key regulators of the PSR. Support here includes: (1) Y1H binding of NAC53 to the PA200 and RPN12a promoters that strongly respond to proteotoxic stress, (2) the ability of NAC53 and NAC78 to homo- and heterodimerize, (3) β-estradiol-induced transcription of an assortment of PSR genes in plants harboring the pEST:NAC78 transgene, and (4) strong transcriptional attenuation of representative PSR genes in plants missing both NAC53 and NAC78. The ability to dimerize is consistent with other NACs that require dimerization to bind a pair of palindromically oriented DNA elements (Olsen et al., 2005b). Connections of NAC53 and/or NAC78 to heat, intense light, drought stress, ROS production, and senescence have also been reported based on the analysis of mutants and overexpression lines (Morishita et al., 2009; Yabuta et al., 2011; Lee et al., 2012, 2014). It is conceivable that these phenotypes are indirectly manifested by direct participation of the pair in proteotoxic stress protection and regulation of 26S proteasome synthesis.

As expected for plants with a dampened PSR, we found that nac53 nac78 double mutants are highly sensitive to proteasome inhibitors. Seedling growth was completely blocked with 50 μM MG132 and 0.5 μM bortezomib for the null nac53-1 nac78-1 plants, with partial inhibition seen for plants harboring the weaker nac53-2 nac78-2 alleles. A similarly modest growth inhibition was seen for the nac78-1 single mutant but not for the single nac53-1 mutant, suggesting that the NAC78 protein is more critical to PSR activation. Although the proteotoxic stress conditions studied here activated expression of NAC53 more robustly than NAC78, we consider it likely that they work simultaneously within the PSR given the similar temporal induction seen for the nac53-1 and nac78-1 single mutants. Regardless of their functions, we note that nac53 nac78 double null mutants are phenotypically normal under nonstress conditions and do not hyperaccumulate Ub conjugates upon MG132 treatment (Supplemental Figure 10). These observations imply that other transcription factors are responsible for the basal synthesis of 26S proteasomes and other protein homeostasis regulators and that during proteotoxic stress, additional recycling pathways are engaged to eliminate the excess Ub conjugates that accumulate if 26S proteasome capacity is insufficient (e.g., autophagy). As an aside, we note that Arabidopsis growth is more sensitive to bortezomib than MG132 and thus might represent a better agent to block 26S proteasome activity in planta.

We consider it likely that NAC53 and NAC78 recognize the same cis-element given their strong amino acid sequence similarity within the NAM DNA binding motif. Unfortunately, the nature of the element is unclear. Prior studies identified the PRCE element with the consensus core TGGGC sequence as the preferred NAC78 binding site (Yabuta et al., 2010; Nguyen et al., 2013), whereas more recent interaction studies with NAC53 and NAC78 reported that both prefers MDM-type motifs (Lee et al., 2012; De Clercq et al., 2013). Understanding this discrepancy will certainly require more in-depth binding studies using NAC53 and NAC78 alone and in combination. Both cis-motifs were common in many of the PSR loci and are present upstream of most 26S proteasome genes. Given the close proximity of PRCE and MDM sequences within proteasome subunit promoters, it is conceivable that the NAC53/NAC78 dimers bind both motifs simultaneously and/or work with other transcription factors that bind.

How NAC53 and NAC78 activate the PSR during proteotoxic stress is unclear. In yeast, their functional counterpart Rpn4 participates in a simple negative feedback circuit whereby Rpn4 becomes stabilized during proteotoxic stress as proteasome capacity becomes overloaded. In mice and humans, its potential counterpart is Nrf1, a transmembrane-containing basic leucine zipper transcription factor that resides on the ER membrane under nonstressed conditions. Upon proteotoxic stress, the DNA binding region of Nrf1 is proteolytically released from the membrane-spanning segment, which then permits its nuclear import to drive 26S proteasome gene expression. Given that NAC53 and NAC78, as part of the Arabidopsis NTL subfamily, are also predicted to have a transmembrane domain, and that others within this subfamily have been reported to use proteolytic release from cytoplasmic stores to regulate their transcriptional activity (Kim et al., 2006; Kim et al., 2007), it is plausible that a shuttle mechanism similar to that of Nrf1 exists. However, this proteolytic step and membrane release, if they occur, might not be essential as the truncated nac53-2 protein described here (Figure 8A), and an engineered truncation of NAC78 by Nguyen et al. (2013) did not appear to constitutively activate the PSR even in the absence of MG132. A similar lack of effect of the C-terminal truncation was seen for another NAC-NTL, NAC13, in its ability to activate transcription (De Clercq et al., 2013), thus questioning the role(s) of the transmembrane domain in NTL-NAC action.

Interestingly, the PSR is populated with transcripts for a number of other NAC transcription factors besides NAC53, including NAC1, 13, 32, 44, 50, 55, 81/ATAF2, 82, and 87, that are robustly upregulated upon MG132 exposure and/or in the rpn10-1 and rpn12a-1 backgrounds. These factors could represent additional transcriptional regulators that assist in the immediate activation of the PSR or reflect part of a transcriptional cascade working downstream that helps activate the full suite of PSR loci. In support of the former, Y2H analyses of these PSR-associated NAC proteins identified NAC13 and NAC81 as NAC53/NAC78 binding partners that could directly participate in PSR activation. NAC13 is particularly intriguing as it, like NAC53 and NAC78, is a member of the NTL subclade and has been implicated in the transcriptional response to mitochondrial dysfunction and binding to MDM sequences (De Clercq et al., 2013). Notably, a number of the gene targets of the mitochondrial dysfunction response are shared with the PSR (e.g., encoding UDP-glycosyl transferases and AOX1a), which combined with the prevalence of MDM sequences in PSR genes, suggests that the two stress responses share a number of protective functions (e.g., oxidative stress defense). A similar overlap between proteotoxic stress and mitochondrial dysfunction has also been observed in mammals (Livnat-Levanon et al., 2014). Taking these results together, an attractive hypothesis is that the PSR loci containing both PRCE and MDM elements are activated by heterodimers bearing NAC13 in combination with either NAC53 or NAC78.

We also note that a subpopulation of PSR genes is likely immune or only weakly responsive to NAC53 and NAC78 regulation (e.g., WRKY25), indicating that other transcription factors are responsible for their activation under proteotoxic stress. One or more of these factors might be found within the PSR (e.g., WRKY6, 25, 33, and 45; Supplemental Figure 2) or present in the list of DNA binding proteins identified as common in our Y1H screen with the PA200 and RPN12a promoters (Supplemental Table 1 and Supplemental Figure 5). Consequently, it is possible that the PSR is controlled by an additional suite of transcriptional regulators beyond NAC53/NAC78 that generate a multilayered system to fully activate the regulon during the various iterations of proteotoxic stress.

METHODS

Plant Materials and Growth Conditions

The T-DNA insertion mutants for NAC53 (nac53-1, SALK_009578C; nac53-2, SALK_018311C) and NAC78 (nac78-1, SALK_025098; nac78-2, SALK_040812C) in the Arabidopsis thaliana ecotype Col-0 (Alonso and Stepanova, 2003) were obtained from the ABRC at Ohio State University (https://abrc.osu.edu/). The rpn10-1 and rpn12a-1 exon-trap lines in the C24 background were generated as previously described (Smalle et al., 2002, 2003). The NAC78 overexpression line (Col-0 background) driven by the β-estradiol-inducible XVE promoter (NAC78 #2138 and #2319) was provided by the TRANSPLANTA resource (Coego et al., 2014).

The proteasome promoter:GUS transgenic lines were generated by PCR amplification of the 5′ upstream region for representative proteasome subunit loci, starting at the end of the upstream coding region (or 2 kb) and terminating at the transcriptional start site. The PCR products were introduced upstream of the full GUS coding region present in the pCAMBIA3301 or pMDC163 vectors. The chimeric genes were transformed into Arabidopsis (Col-0 ecotype) by the floral dip method using the Agrobacterium tumefaciens strain GV3101 (Lee et al., 2011). Basta-resistant seedlings were screened for GUS activity by histochemical staining with the substrate X-Gluc (5-bromo-4-chloro-3-indolyl-β-glucuronic acid; Sigma-Aldrich). For quantitative studies, total extracts from 10-d-old seedlings were assayed for GUS activity using the fluorescence-based MUG assay (Sigma-Aldrich; Lee et al., 2011). At least 30 independent transformants were examined for each construction to avoid artifacts generated by the site of the transgene insertion. For examination of GUS staining patterns, 6-d-old seedlings were incubated overnight in X-Gluc following a 12-h exposure to 100 μM MG132 dissolved in DMSO, using an equivalent volume of DMSO as the control.

Unless otherwise noted, seeds were surface sterilized, stratified in the dark at 4°C for 2 d, and then germinated on 0.7% agar containing half-strength Murashige and Skoog (MS) medium (Caisson Labs), 1% sucrose, and 0.5% MES (pH 5.7), with or without various concentrations of MG132 [N-(benzyloxycarbonyl)-leucinyl-leucinyl-leucinal; SelleckChem] or bortezomib [(1R)-3-methyl-1-(((2S)-3-phenyl-2-(pyrazin-2-carbonylamino) propanoyl)amino)butyl)boronic acid; SelleckChem]. For the RNA-seq studies, the seedlings were grown for 5 d in 12-well liquid cultures plates under continuous fluorescent white light at 22°C prior to RNA isolation. For studies on the phenotypic effects of MG132 or bortezomib, stratified seeds were plated on the same medium with the addition of 0.7% agar. After 6 to 10 d growth under a 16-h-light/8-h-dark photoperiod at 22°C, the plants were weighed individually or in batches of two to five seedlings to determine fresh weight. One-way ANOVA was used to determine statistical significance among the various genetic backgrounds and treatments.

RT-qPCR and RNA-Seq Analyses

Following various treatments, liquid-grown seedlings were pressed dry, frozen in liquid nitrogen, and pulverized. Total RNA was extracted with QIAzol lysis reagent (Qiagen), treated with DNaseI (Promega), and then converted to cDNA using Superscript III reverse transcriptase (Life Technologies). RT-qPCR amplifications on the cDNA populations were performed with a Roche LightCycler 480 using either Roche LightCycler 480 SYBR Green Master Mix or MidSci Bullseye SYBR Green Master Mix. Appropriate priming sites for each locus were identified using Primer3Plus (http://www.bioinformatics.nl/cgi-bin/primer3plus/primer3plus.cgi). See Supplemental Table 2 for the list of primers. Primer efficiencies were experimentally determined to be between 1.90 and 2.10, based on assay of a standard dilution series (Marshall et al., 2015). The relative abundance of each transcript was determined by the comparative threshold cycle method (Pfaffl, 2001) using the ACT2 and PP2A reference genes as internal controls. All data were normalized to untreated wild type.

mRNA enrichment and library generation for RNA-seq studies were performed using the TruSeq RNA library sample preparation kit v2 (Illumina) with help from the University of Wisconsin-Madison Gene Expression Center (http://www.biotech.wisc.edu/services/gec). Multiplexed sequencing was performed on the Illumina HiSeq 2500 platform using 100-bp single-ended reads. At least two biological replicates were analyzed for each condition with each amplification yielding between 10 and 25 million raw sequencing reads. FASTQ files were quality checked with the Trimmomatic package (Bolger et al., 2014). Between 70 and 80% of reads were retained in all biological replicates. Trimmed reads were aligned by TopHat-Bowtie 2 to the TAIR 10 genome using the Ensemble 2013 Arabidopsis.gtf file annotations (Langmead and Salzberg, 2012). Read counts were generated through HTSeq (Anders et al., 2015), quantified via edgeR (Robinson et al., 2010), and then normalized to the wild-type Col-0 untreated control. Only those transcripts that met a FDR ≤ 0.05 cutoff were included in the analyses.

All available gene coexpression and protein interaction data were collected from the STRING database (Jensen et al., 2009) and mapped using the organic layout option from Cytoscape 3.0.1 (Shannon et al., 2003). Gene Ontology and functional term enrichments were determined using DAVID (Huang et al., 2009). Highly interconnected gene node clusters were identified using the MCODE plug-in for Cytoscape (Bader and Hogue, 2003). Heat maps of significantly expressed genes were generated by Java TreeView v1.1643 (Saldanha, 2004).

Immunoblot Analysis

Immunoblot analyses were conducted with 6-d-old seedlings homogenized directly into SDS-PAGE sample buffer. Following SDS-PAGE, proteins were transferred onto Millipore Immobilon-P or Immobilon-FL membranes and probed with antibodies against PA200 (Book et al., 2010), Ub (van Nocker et al., 1996), RPN1, RPN5, RPN10, RPN12a, and RPT2a (Smalle et al., 2002; Yang et al., 2004), and histone H3 (AbCam; AB1791).

Phylogenetic Comparisons of NAC Protein Sequences

Amino acid sequences were aligned by MEGA v6.0 under the MUSCLE default settings (Tamura et al., 2013). Alignments are provided in Supplemental Data Sets 5 and 6. The locations of the possible NAM and membrane spanning regions were predicted by PFAM v28.0 (http://pfam.xfam.org). Phylogenetic analyses of the predicted full-length NAC proteins or only their NAM DNA binding regions from Arabidopsis and other plant species available in Phytozome (www.phytozome.jgi.doe.gov/) or NCBI (www.ncbi.nlm.nih.gov/genbank) were performed using MrBayes 3.2 (Ronquist et al., 2012) and the mixed amino acid model (aamodelpr = mixed) until convergence (average sd of split frequencies) reached below 0.05 or plateaued. The trees were rooted with the Physcomitrella patens NAC protein gi_168025227 as the outgroup and visualized using FigTree v1.4.2. Nucleotide sequences related to the consensus PRCE and MDM cis-motifs (De Clercq et al., 2013; Nguyen et al., 2013) were detected in the region upstream of the translation start site using the MEME sequence enrichment identification algorithm with a 7th-order background model calculated from all Arabidopsis intergenic sequences (Bailey et al., 2006).

BiFC, Y1H, and Y2H Analyses

For BiFC, the full-length NAC53 or NAC78 coding sequences in the pDONR221 plasmid were introduced into either the pSITE-N-EYFP-C1 or pSITE-C-EYFP-C1 vectors (ARBC stock numbers CD3-1648 [nYFP] or CD3-1649 [cYFP], respectively), and transformed into the Agrobacterium tumefaciens strain GV3101 (Li et al., 2014). Overnight cultures were diluted to OD₆₀₀ of 0.5 in resuspension buffer (10 mM MgCl₂, 10 mM MES, pH 5.7, and 100 μM acetosyringone) and syringe-infiltrated into 4- to 6-week-old Nicotiana benthamiana leaves. Fluorescence within the infiltrated regions was visualized after 36 h using a Zeiss 510 Meta confocal laser scanning microscope. For MG132 treatments and 4′,6-diamidino-2-phenylindole (DAPI) staining, the resuspension buffer without acetosyringone and containing 100 μM MG132 and/or 1 μM DAPI was infiltrated into the same leaf regions 24 or 1.5 h prior to visualization, respectively.

Y2H assays were performed using the ProQuest Two-Hybrid System (Life Technologies). The indicated NAC genes were amplified by PCR from Arabidopsis cDNA generated as described above and recombined into pDONR221 via the Gateway BP clonase II reaction. These fragments were then recombined in-frame to either the GAL4 activation domain or GAL4 binding domain coding sequences in the pDEST22 or pDEST32 vectors (Life Technologies). Constructs were verified by sequencing, and pairwise combinations of genes in pDEST22 and pDEST32 (or the empty vectors as controls) were cotransformed into the Saccharomyces cerevisiae strain MaV203 (Vidal et al., 1996). Y2H assays were performed by diluting overnight yeast cultures with either YPD medium lacking Leu and Trp or lacking Leu, Trp, and His and containing the indicated concentrations of 3-AT, and then growing the yeast for 2 d at 30°C on these media for nonselective and selective growth, respectively.

For a nonbiased Y1H screen, 1000- and 253-bp fragments of the PA200 and RPN12a upstream regions, respectively, were cloned into a pLacZ plasmid and screened individually against 1700 Arabidopsis transcription factors cloned into the MaV203 yeast strain (Gaudinier et al., 2011). For confirmatory Y1H screens, lacZ activation was quantified using the Miller β-galactosidase assay and the substrate 5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside as previously described (Zhang and Bremer, 1996). At least three biological replicates were averaged for each measurement.

Accession Numbers

Sequence data from this article can be found in the Arabidopsis Genome Initiative or GenBank/EMBL databases under the following accession numbers: ACT2 (At3g18780), CDC48a (At3g09840), GSTU25 (At1g17180), HSF8A (At1g67970), NAC1 (At1g01010), NAC13 (At1g32870), NAC32 (At1g77450), NAC44 (At3g01600), NAC50 (At3g10480), NAC53 (At3g10500), NAC55 (At3g15500), NAC78 (At5g04410), NAC81 (At5g08790), NAC82 (At5g09330), NAC87 (At5g18270), NAS6 (At2g03430), PA200, (At3g13330), PBA1 (At4g31300), PP2A (At1g13320), RPN1a (At2g20580), RPN1b (At4g28470), RPN3a (At1g20200), RPN3b (At1g75990), RPN5a (At5g09900), RPN5b (At5g64760), RPN10 (At4g38630), RPN12a (At1g64520), RPT1a At1g53750), RPT1b (At1g53780), RPT2a (At4g29040), RPT2b (At2g20140), RPT4a (At5g43010), RPT4b (At1g45000), UFD1 (At2g21270), and WRKY25 (At2g30250). The RNA-seq data are available in the NCBI Gene Expression Omnibus under accession number GSE81668.

Supplemental Data

• Supplemental Figure 1. RNA-Seq Comparison of Proteotoxic Stress on the Expression of Several Transcripts Commonly Used as Standards for RT-qPCR Studies.

• Supplemental Figure 2. Expression Heat Maps of Representative PSR Genes Outside of Those Encoding 26S Proteasome Subunits.

• Supplemental Figure 3. Description of the 24-h MG132 Upregulated Gene Interaction and Coexpression Network.

• Supplemental Figure 4. PRCE and MDM Sequences Are Located in the Upstream DNA Region of 26S Proteasome Subunits.

• Supplemental Figure 5. Occurrence of AGRIS-Annotated DNA Binding Motifs in the Upstream Region of 26S Proteasome and Other PSR Genes.

• Supplemental Figure 6. Phylogenetic Tree of the 109 Arabidopsis NAC Transcription Factors Using the Full Amino Acid Sequences for Comparison.

• Supplemental Figure 7. Phylogenetic Tree of the 109 Arabidopsis NAC Transcription Factors Using just the NAM Domain Sequences for Comparison.

• Supplemental Figure 8. Amino Acid Sequence Alignment of NTL Proteins NAC53, NAC78, and NAC13.

• Supplemental Figure 9. Bimolecular Fluorescence Complementation of NAC53 and NAC78 with or without MG132 Treatment.

• Supplemental Figure 10. Ub Conjugate Levels in nac53-1 and nac78-1 Mutants with or without MG132 Treatment.

• Supplemental Figure 11. The PSR in Single nac53 and nac78 Mutants Has the Same Temporal Response to MG132 as the Wild Type.

• Supplemental Table 1. List of Y1H Prey in Common When Using the Arabidopsis PA200 and RPN12a Promoters as Bait.

• Supplemental Table 2. List of Oligonucleotide Primers Used for RT-qPCR Analysis.

• Supplemental Data Set 1. List of Genes Whose Expression Was Significantly Affected by 3-h Exposure to MG132.

• Supplemental Data Set 2. List of Genes Whose Expression Was Significantly Upregulated in the rpn10-1 and rpn12a-1 Mutant Backgrounds.

• Supplemental Data Set 3. List of Genes within the PSR.

• Supplemental Data Set 4. List of Genes Whose Expression Was Significantly Affected by Long-Term Exposure to MG132.

• Supplemental Data Set 5. Text File of the Full NAC Protein Alignment Corresponding to the Phylogenetic Analysis in Supplemental Figure 6.

• Supplemental Data Set 6. Text File of the NAM Domain Alignment Corresponding to the Phylogenetic Analysis in Supplemental Figure 7.

Glossary

RP

regulatory particle

CP

core protease

PSR

proteasome stress regulon

ER

endoplasmic reticulum

MUG

4-methylumbelliferyl-β-d-glucuronide

FDR

false discovery rate

MDM

mitochondrial dysfunction motif

Y1H

yeast one-hybrid

ROS

reactive oxygen species

Y2H

yeast two-hybrid

3-AT

3-amino-1,2,4-triazole

BiFC

bimolecular fluorescence complementation

DAPI

4′,6-diamidino-2-phenylindole