Quality control (QC) of gene expression is important for protein homeostasis or proteostasis. QC machineries repair or degrade defective proteins to maintain the integrity of cellular proteome (Balchin et al., 2016). The QC process starts in the nucleus where transcription and RNA processing are monitored. As a result, only mature mRNA exits the nucleus. The process continues in the cytoplasm where several QC mechanisms target defective RNAs and proteins. During the first round of translation, aberrant mRNA is recognized and degraded by mRNA surveillance-triggered decay pathways, including the nonsense-mediated decay pathway (Chantarachot and Bailey-Serres, 2018). Upon translational repression, mRNA can be either directly degraded by cytosolic nucleases or sequestered in cytoplasmic compartments containing ribonucleoproteins (RNPs), such as stress granules (SGs) and processing bodies (Chantarachot and Bailey-Serres, 2018)(Fig. 1). SGs contain mRNA, RNA binding proteins, translation factors, and 40S ribosome subunits. SG assembly is induced by stresses, such as heat shock and hypoxia. When cells recover from stress, messenger RNP (mRNP) can be released from SGs and participate in translational reinitiation. Thus, SGs are considered a reservoir of stalled translation initiation complexes that accumulate during global repression of translation induced by stress. Processing bodies contain mRNA decay factors in addition to some SG components. However, mRNAs associated with processing bodies appear stable, and the biological role of processing bodies is still enigmatic (Chantarachot and Bailey-Serres, 2018).

Protein QC requires collaboration of three systems that are conserved in yeast and metazoans – the chaperone network, the ubiquitin-proteasome system (UPS), and autophagy (Balchin et al., 2016; Harper and Bennett, 2016). Molecular chaperones assist correct folding of nascent and misfolded proteins. When proteostasis is not maintained because of heat shock and other proteotoxic stresses, misfolded proteins build up and form various types of protein aggregates (Fig. 1). To avoid the potential toxicity of such aggregates, cells may sequester them in compartments specialized for protein QC, such as the aggresome in mammals (Sontag et al., 2017). The chaperone network releases individual polypeptides from protein aggregates and refolds them (Mogk et al., 2018). A portion of misfolded polypeptides is not repaired and can be tagged with ubiquitin for degradation by the UPS. Alternatively, QC compartments sequestering protein aggregates can be enclosed by the autophagosome, which fuses with the vacuole, or lysosome in metazoans, where its contents are degraded (Fig. 1).

Recent studies indicate that plant QC system for cytoplasmic proteins also consists of chaperone network (Wang et al., 2004), UPS, and autophagy (Kim et al., 2017; Shen et al., 2007; Zhou et al., 2014). This review will focus on the role of plant autophagy in the QC of defective proteins and RNPs that may form membrane-less compartments in the cytoplasm. We will consider a mechanism by which these structures are recognized and sequestered by autophagic vesicles. We will not discuss selective autophagy for membranous organelles in plants, as its roles in the QC of proteins targeted to the endoplasmic reticulum (ER)(Strasser, 2018), peroxisomes (Young and Bartel, 2016), plastids (Nakamura and Izumi, 2018; Otegui, 2018), and mitochondria (Broda et al., 2018) have been discussed in previous reviews.

Autophagy is a membrane trafficking route by which cytoplasmic constituents are sequestered by the autophagosome and targeted to the vacuole or lysosome for bulk degradation. In yeast and higher eukaryotes, autophagy is mediated by a conserved set of core Autophagy-related (ATG) proteins (Marshall and Vierstra, 2018a). During starvation, nutrients recycled from autophagic degradation of cytoplasmic materials are important for survival. Starvation-induced autophagy typically sequesters a portion of the cytoplasm without selectivity, although the phagophore, the precursor of the autophagosome, does not seem to be initiated at a random location. Under a nutrient-sufficient condition, cells maintain a basal level of autophagy, enough to eliminate obsolete and defective components. This housekeeping function is often achieved by selective autophagy (Marshall and Vierstra, 2018a), in which the phagophore selectively sequesters its cargo. Generally, autophagic cargo interacts either with core ATG proteins (especially ATG8 covalently attached to autophagic membranes) or with autophagy receptors, which in turn bind to ATG8 (Marshall and Vierstra, 2018a). Cargo is often ubiquitylated and recognized by autophagy receptors that contain one or more domains with affinity for ubiquitin, in addition to the ATG8-interacting motif (AIM). In mammals, Atg8 and ubiquitylated cargo are connected by various autophagy receptors, including p62 and p62-related NBR1. Like many other autophagy receptors, p62 and NBR1 remain associated with cargo and ATG8 until they are finally degraded (Fig. 1). Many plant species have only one homolog for p62 and NBR1 (Svenning et al., 2011; Zientara-Rytter et al., 2011). Arabidopsis NBR1 also binds ubiquitin in vitro, interacts with ATG8, and is delivered to the vacuole via an autophagic trafficking pathway (Svenning et al., 2011).

Mammalian p62 and NBR1 mediate the selective autophagy of protein aggregates, termed aggrephagy (Danieli and Martens, 2018; Dikic, 2017). Aggrephagy in plant cells was described in tobacco cells expressing red fluorescent proteins fused to ER-localized proteins (Toyooka et al., 2006). The fusion proteins were mislocalized in the cytosol and detected as small puncta which were delivered to the vacuole via ATG8-positive autophagic vesicles (Toyooka et al., 2006). In heat-exposed Arabidopsis plants, mutations in either NBR1 or core ATG genes resulted in the over-accumulation of ubiquitylated insoluble proteins (Zhou et al., 2013). Thus, NBR1 may have a role in aggrephagy during heat stress. Whether plant NBR1 mediates the formation of autophagosomes containing protein aggregates remains to be determined by microscopic analysis using aggregation-prone protein reporters.

RNP compartments are also degraded by selective autophagy. In yeast, SGs and processing bodies are delivered to the vacuole during autophagy, and autophagic clearance of yeast and mammalian SGs requires the AAA-ATPase CDC48/VCP (Buchan et al., 2013). It is unknown whether plant SGs and processing bodies are degraded by autophagy. Selective autophagy of cytoplasmic RNP granules, termed granulophagy, is evident during proteotoxic stress and appears to be linked to other components of the protein QC system (Fig. 1). In yeast, heat-inducible SGs behave like solid, reminiscent of protein aggregates, and chaperones promote SG disassembly. Mammalian SGs normally behave like liquid droplets, but mutations encoding misfolding-prone proteins can result in the formation of aberrant, solidified SGs that accumulate aggregated proteins called defective ribosomal products. The aberrant SGs are targeted to the aggresome and degraded by selective autophagy, although disaggregation and refolding of misfolded SG proteins by chaperones are preferred over autophagic clearance (Ganassi et al., 2016; Mateju et al., 2017).

Some polypeptides are assembled to form large protein complexes with a molecular mass of > 1 MDa, such as ribosomes, spliceosomes, and proteasomes. Chaperones and specific UPS factors are needed when cells repair or selectively degrade defective components of protein complexes (Juszkiewicz and Hegde, 2018). Nevertheless, autophagic degradation of defective ribosomes and proteasomes en bloc can be advantageous. When cells undergo nutrient stress, not all ribosomes and proteasomes are needed and their excess can be recycled to supply amino acids and other breakdown products. In other cases, severe proteome imbalance and proteotoxic stress can result in the accumulation of defective ribosomes and proteasomes, which are toxic unless they are quickly eliminated by selective autophagy.

Selective autophagy of proteasomes, or proteaphagy, is mediated by various receptors, specifically, RPN10 in Arabidopsis (Marshall et al., 2015), Cue5 in yeast (Marshall et al., 2016), and p62 in mammals (Cohen-Kaplan et al., 2016). In both Arabidopsis and yeast, proteaphagy is activated upon nitrogen starvation and proteasome inhibition. When proteasome is inhibited, proteasome subunits are ubiquitylated and recognized by the proteaphagy receptors (Marshall et al., 2015; 2016). Like p62, Cue5 is also known as aggrephagy receptors (Lu et al., 2014). Hsp42, the oligomeric chaperone found in protein QC compartments in yeast, is required for the aggregation of defective proteasomes (Marshall et al., 2016). These findings imply that proteaphagy and aggrephagy have a common mechanism for cargo recognition. Interestingly, proteaphagy in Arabidopsis is not stimulated by carbon starvation, although carbon starvation still induces non-selective autophagy. In this case, functional proteasomes are sequestered in proteasome storage granules that presumably prevent them from becoming autophagic cargo (Marshall and Vierstra, 2018b). Based on this collection of observations, it is suggested that protein QC in plant cells requires compartmentalization and collaboration of chaperones, the UPS, and autophagic machinery.

Nitrogen starvation also induces selective autophagy of ribosomes, or ribophagy, in yeast (Kraft et al., 2008) and mammalian cells (Wyant et al., 2018). Yeast mutants defective in ribophagy had a high level of ubiquitylation of ribosomal proteins, suggesting a role of ubiquitylation in cargo selection during starvation-induced ribophagy (Kraft et al., 2008). Nuclear fragile X mental retardation-interacting protein 1 (NUFIP1) was identified as a mammalian ribophagy receptor that interacts with a mammalian homolog of ATG8 and with the 60S subunit of the ribosome (Wyant et al., 2018), although a ribosome-interacting motif has yet to be mapped in NUFIP1. In plants, the vacuolar ribonuclease RNS2 is responsible for rRNA degradation via constitutive autophagy (Floyd et al., 2017; Hillwig et al., 2011). rns2 mutants show defective recycling of rRNA and activate compensatory mechanisms, such as autophagy induction (Floyd et al., 2015) and metabolic change for nucleoside biosynthesis (Morriss et al., 2017). Although direct evidence for ribophagy in plants is lacking, it is notable that Arabidopsis NUFIP protein (Rodor et al., 2011) is predicted to have an AIM (http://repeat.biol.ucy.ac.cy/iLIR/).

Two kinds of interactions regulate cargo selection: direct interaction with ATG8 or other core ATG proteins and indirect interactions with autophagy receptors. These interactions are often regulated not only by post-translational modification of interacting interfaces but also by receptor oligomerization.

A few plant autophagic cargo proteins appear to directly interact with ATG8 and other ATG proteins. For example, components of the ATG1 kinase complex in Arabidopsis (Li et al., 2014; Suttangkakul et al., 2011) that contain AIMs are degraded in the vacuole during starvation-induced autophagy. Another ATG8-interacting cargo is a virulence factor of Cotton leaf curl Multan virus (Haxim et al., 2017). The RNA-dependent RNA polymerase of Turnip mosaic virus is autophagic cargo that interacts with tobacco ATG6, one of the core ATG proteins in an autophagy regulatory complex (Li et al., 2018).

Ubiquitylation of cargo for aggrephagy and other types of selective autophagy in yeast and metazoans is common, although ubiquitylation-independent selection is possible (Grumati and Dikic, 2018). In Arabidopsis, proteasome subunits were ubiquitylated and degraded by proteaphagy during proteasome inhibition (Marshall et al., 2015). The proteaphagy receptor RPN10 interacts with ubiquitylated proteasome subunits through its ubiquitin-interacting motif. The mammalian proteaphagy receptor p62 also interacts with ubiquitylated proteasome subunits during amino acid starvation (Cohen-Kaplan et al., 2016). Although p62-related Arabidopsis NBR1 binds ubiquitin chains (Svenning et al., 2011), it is not required for proteaphagy in Arabidopsis (Marshall et al., 2015).

The transcription factor BRI1-EMS Suppressor 1 (BES1), involved in brassinosteroid signaling, was also identified as autophagic cargo that is ubiquitylated (Nolan et al., 2017). Ubiquitylated BES1 interacts with DSK2A, an autophagy adaptor containing a ubiquitin-like domain, two AIMs, and a ubiquitin-associated (UBA) domain. Cytoplasmic BES1 colocalizes with ATG8 and is degraded in the vacuole in a DSK2- and core ATG-dependent manner (Nolan et al., 2017). Eukaryotic DSK2 homologs are encoded by the single gene Dsk2 in yeast, two genes (DSK2A and DSK2B ) in Arabidopsis, and four genes (Ubiquilin1 to 4 ) in humans. Whereas Ubiquilin4 is involved in autophagy (Lee et al., 2013), yeast Dsk2 delivers ubiquitylated proteins to the proteasome (Zhang et al., 2009). Interestingly, Ubiquilin2 is required for the proteasomal degradation of ubiquitylated and misfolded proteins in the nucleus (Hjerpe et al., 2016; Samant et al., 2018). It is unclear whether Arabidopsis DSK2B plays a similar role in nuclear protein QC, but DSK2B interacts with RPN10, a proteasome subunit (Lin et al., 2011). Furthermore, bimolecular fluorescence complementation experiments indicated that DSK2B interacts with BES1 in a nucleus-like compartment, whereas DSK2A-BES1 interaction is detected as diffuse and punctate signals in the cytoplasm (Nolan et al., 2017). BES1 stabilization and brassinosteroid response phenotypes are more prominent in DSK2A/B double RNAi lines than in autophagy-defective mutants (Nolan et al., 2017). These data are consistent with our proposal that DSK2A and DSK2B function as selective autophagy adaptors and/or UPS shuttle factors (Fig. 2). Arabidopsis SINAT family proteins are E3 ligases that ubiquitylate BES1 for its light-dependent degradation (Yang et al., 2017). All five SINAT family proteins were localized in both the nucleus and cytoplasm (Yang et al., 2017), where SINAT-mediated ubiquitylation of BES1 likely takes place. What is the biological significance of autophagic degradation of excess BES1? Autophagy may downregulate the capacity of cytoplasmic BES1 reservoir, thereby preventing re-entry of BES1 into the nucleus and assisting in the termination of brassinosteroid signaling (Fig. 2).

Ubiquitylated proteins sequestered in mammalian QC compartments may be disaggregated and either refolded by chaperones or degraded by the proteasome. Alternatively, the whole QC compartments may be targeted to the vacuole via the autophagy pathway (Fig. 1). The relationship between ubiquitylation of proteasomal substrates and the ubiquitylation of aggrephagy cargo is not fully understood (Kwon and Ciechanover, 2017). However, increasing evidence indicates that ubiquitin-based selection of autophagic cargo can be achieved by multi-valent, weak interactions between polyubiquitin chains on the cargo surface and ubiquitin-binding domains of autophagy receptors. The interactions are enhanced by intermolecular interactions between autophagy receptors. Mammalian p62 self-oligomerizes and forms filamentous structures, which interact with ubiquitylated proteins. It has been proposed that these biochemical properties of p62 promote controlled protein aggregation. More recently, liquid-liquid phase separation (LLPS) has been proposed as a mechanism for concentrating autophagic cargo molecules (Danieli and Martens, 2018). LLPS is responsible for the formation of biomolecular condensates, which include several membrane-less compartments, specifically nucleoli, Cajal bodies, SGs, processing bodies, and protein QC compartments. These condensates rapidly exchange their constituents with surrounding environment and have liquid-like physical properties, analogous to oil droplets separating from surrounding water. Depending on changes in condensate composition and external variables (for example, ionic strength, pH, and temperature), liquid-like condensates can turn into gel-like compartments. Condensates that are p62-positive can grow in vitro to a few micrometers in size when ubiquitin chains are added (Sun et al., 2018; Zaffagnini et al., 2018). In p62 condensates, the p62 filament likely provides a scaffold to which ubiquitylated misfolded proteins can adhere (Danieli and Martens, 2018).

Although the composition of RNP granules such as SGs and processing bodies in plant cells is similar to those in yeast and metazoans (Chantarachot and Bailey-Serres, 2018), their properties as membrane-less condensates have yet to be tested (Cuevas-Velazquez and Dinneny, 2018). Protein QC compartments in plant cells and their relationship to RNP granules are not well understood, partly because amenable models of aggregate-forming proteins are scarce in plants. However, putative protein QC compartments in Arabidopsis were induced upon proteasome inhibition and co-localized with chaperones and proteasomes (Oh et al., 2017). It will be interesting to know more about the putative QC compartments, particularly: if they have the liquid-like properties of condensates, what the requirements for their assembly are, and if autophagy is involved in their clearance.

Important questions remain in the study of selective autophagy for protein QC. Where and how do phagophore nucleation and expansion occur during selective autophagy? What is the source of lipid for phagophore expansion? What is the mechanism for autophagic cargo selection favoring terminally misfolded aggregates that cannot be handled by chaperones and the proteasome?

We also have specific questions for the autophagy of plant QC compartments. What functions are shared by plant NBR1 and mammalian p62 and NBR1? Does DSK2A use a p62-like mechanism to mediate the autophagy of ubiquitylated BES1 and possibly other unknown cargo? Do plants have additional autophagy receptors for protein and RNA QC? A panel of protein QC substrates for plants will be useful to answer these questions.

It has been proposed that the sessile nature of plants demands a high capacity of plant QC systems to confront inevitable environmental stress. Better understanding of RNA and protein QC systems in plants will help plant biotechnologists improve stress tolerance of crops as well as increase yield in molecular farming, which often involves transgenic expression of unstable heterologous proteins.

I apologize to colleagues whose work has not been mentioned because of space limitations. This work is supported by grants NRF-2017R1A2B4002335 to T.C.