ML3 is a NEDD8- and ubiquitin-modified protein.

A previously uncharacterized lipid-binding protein is modified by the ubiquitin-like protein NEDD8 and by ubiquitin. NEDD8 (NEURAL PRECURSOR CELL-EXPRESSED, DEVELOPMENTALLY DOWN-REGULATED PROTEIN8) is an evolutionarily conserved 8-kD protein that is closely related to ubiquitin and that can be conjugated like ubiquitin to specific lysine residues of target proteins in eukaryotes. In contrast to ubiquitin, for which a broad range of substrate proteins are known, only a very limited number of NEDD8 target proteins have been identified to date. Best understood, and also evolutionarily conserved, is the NEDD8 modification (neddylation) of cullins, core subunits of the cullin-RING-type E3 ubiquitin ligases that promote the polyubiquitylation of degradation targets in eukaryotes. Here, we show that Myeloid differentiation factor-2-related lipid-recognition domain protein ML3 is an NEDD8- as well as ubiquitin-modified protein in Arabidopsis (Arabidopsis thaliana) and examine the functional role of ML3 in the plant cell. Our analysis indicates that ML3 resides in the vacuole as well as in endoplasmic reticulum (ER) bodies. ER bodies are Brassicales-specific ER-derived organelles and, similar to other ER body proteins, ML3 orthologs can only be identified in this order of flowering plants. ML3 gene expression is promoted by wounding as well as by the phytohormone jasmonic acid and repressed by ethylene, signals that are known to induce and repress ER body formation, respectively. Furthermore, ML3 protein abundance is dependent on NAI1, a master regulator of ER body formation in Arabidopsis. The regulation of ML3 expression and the localization of ML3 in ER bodies and the vacuole is in agreement with a demonstrated importance of ML3 in the defense to herbivore attack. Here, we extend the spectrum of ML3 biological functions by demonstrating a role in the response to microbial pathogens.

The 8-kD protein ubiquitin is a well-studied modifier of eukaryotic proteins that is best known for targeting proteins conjugated to Lys-48-linked ubiquitin chains for degradation by the 26S proteasome (Komander and Rape, 2012). In addition, Lys-63-linked ubiquitin chains are required for targeting membrane proteins for degradation to the vacuole or the lysosome, and monoubiquitylation as well as polyubiquitylation events have been shown to control the activity, fate, or cellular behavior of proteins (Komander and Rape, 2012). E3 ubiquitin ligases recognize the ubiquitylation targets and promote ubiquitylation, while deubiquitylating enzymes are able to hydrolyze ubiquitin linkages (Hotton and Callis, 2008;Deshaies and Joazeiro, 2009;Komander et al., 2009). In addition to ubiquitin, several other ubiquitin-related proteins regulate cellular functions in eukaryotic cells, such as SMALL UBIQUITIN-LIKE MODIFIER (SUMO) or RELATED TO UBIQUITIN/NEURAL PRECURSOR CELL-EXPRESSED, DEVELOPMENTALLY DOWN-REGULATED (RUB/NEDD8 [herein NEDD8]; Rabut and Peter, 2008;Praefcke et al., 2012;Vierstra, 2012).
The fact that NEDD8 is the closest homolog of ubiquitin suggests that NEDD8 may have a similarly broad range of targets and activities as ubiquitin. However, only a very limited number of NEDD8-modified proteins has been identified to date (Rabut and Peter, 2008;Xirodimas, 2008;Ma et al., 2013). Best understood is the role of neddylation in regulating the cullin subunits of cullin-RING-type E3 ubiquitin ligases (Duda et al., 2008). Eukaryotic cells contain different types of cullin-RING ligases of varying architecture and varying substrate specificity (Duda et al., 2011;Harper and Tan, 2012). Cullin neddylation promotes E3 ligase complex activity as well as ubiquitylation by controlling E3 assembly and by inducing conformational rearrangements that promote substrate ubiquitylation (Duda et al., 2008). Additionally, a small number of animal proteins have been described as NEDD8 targets in recent years, but the biologically significance of their NEDD8 modification is only vaguely understood (Xirodimas et al., 2004Rabut and Peter, 2008;Mahata et al., 2012;Noh et al., 2012;Ma et al., 2013).
ML proteins are defined as proteins with an MD-2related lipid-recognition domain (Inohara and Nuñez, 2002). MD-2 is an extracellular binding partner of Toll-LIKE RECEPTOR4 (TLR4), and together, the two proteins mediate the response to bacterial lipopolysaccharides that are recognized as pathogen-associated molecular patterns that induce innate immunity responses in mammals (Viriyakosol et al., 2001;Kim et al., 2007;Park et al., 2009). The crystal structure of MD-2 has been resolved in its complex with TLR4 (Park et al., 2009). MD-2 consists of two antiparallel b-sheets that form a hydrophilic pocket for binding of the lipopolysaccharide ligand (Park et al., 2009). The protein Niemann-Pick-type C2 (NPC2) is another well-characterized ML protein. NPC2 binds cholesterol in the mammalian lysosome to initiate the transport of cholesterol across the lysosomal membrane via the activity of the transporter NPC1 (Xu et al., 2007). Loss of NPC1 and NPC2 leads to NPC disease, a rare lipid storage disorder where intracellular lipid transport is disrupted, leading to the accumulation of lipid products in late endosomes and lysosomes. The biochemical and cell biological functions of plant ML proteins are completely unresolved.
The endoplasmic reticulum (ER) exhibits a variety of shapes and movements in eukaryotic cells. The ER is composed of an extensive network of cisternae and tubules and can give rise to a range of ER-derived compartments that vary in size from 0.1 to 10 mm. ER bodies are such ER-derived compartments that were originally observed by electron microscopy in radish (Raphanus sativus) root cells as organelles of unknown origin (Bonnett and Newcomb, 1965). In Arabidopsis (Arabidopsis thaliana), these plant-specific spindle-shaped organelles were first recognized as ER-derived structures based on their ability to retain the ER marker protein green fluorescent protein (GFP)-HDEL, which carries an ER-targeting signal as well as an ER-retention signal (Haseloff et al., 1997;Ridge et al., 1999;Hayashi et al., 2001). In Arabidopsis, ER bodies can be found in seedlings and mature roots, and their formation can be induced in mature rosette leaves by methyl jasmonate (MeJA), a phytohormone induced in response to herbivore attack as part of the internal defense response and for signaling to neighboring plants. Mutant screens for ER body-deficient mutants have so far led to the identification of two proteins required for ER body formation, NAI1 and NAI2 (Matsushima et al., 2004;Yamada et al., 2008Yamada et al., , 2009). NAI1 is a basic helix-loop-helix transcription factor essential for the formation of ER bodies in seedlings and roots that regulates the expression of the ER body proteins PYK10/BGLU23 and NAI2 (Matsushima et al., 2004). NAI2 is an ER body protein of unknown function with 10 glutamic acidphenylalanine-glutamic acid repeats that has a signal peptide but, interestingly, no ER retention signal (Yamada et al., 2008. The purification of ER bodies from roots followed by mass spectrometric analysis resulted in the identification of the b-glucosidases PYK10/BGLU23 and BGLU21 as two major components of ER bodies Nagano et al., 2008). The abundance of PYK10/BGLU23 correlates in different stages of plant growth and development with the absence, induction, and presence of ER bodies, and mutations in PYK10/BGLU23 and BGLU21 affect ER body size positively . PYK10/BGLU23, therefore, is thought to be a major and possibly specific component of ER bodies.
At present, ER bodies have only been observed in species of the order Brassicales, which includes Arabidopsis and Brassica rapa. The observation that ER bodies are induced by jasmonate has given rise to the hypothesis that ER bodies may participate in plant-pathogen responses (Matsushima et al., 2002(Matsushima et al., , 2004Hara-Nishimura and Matsushima, 2003). ER bodies may form in response to pathogen attack to release hydrolytic enzymes to fend off the herbivore after wounding (e.g. by hydrolyzing inactive secondary metabolites such as scopolin into active scopoleptin; Ahn et al., 2010). In support of this hypothesis, a deletion in the NAI1 promoter was associated with increased susceptibility to the mutualistic fungus Piriformospora parasitica (Sherameti et al., 2008). Jacalinrelated lectins and GDSL lipase-like proteins may contribute to such defense responses by forming complexes with PYK10/BGLU23. Also, ML3 has recently been linked to defense signaling in a study that showed that ml3 mutants are hypersensitive to herbivore attack (Fridborg et al., 2013).
In this study, we characterize the ML domain protein ML3. We and others have previously identified ML3 as a putatively NEDD8-modified protein (Hakenjos et al., 2011;Hotton et al., 2012). Here, we show that ML3 is indeed a NEDD8-as well as a ubiquitin-conjugated protein in planta. ML3 is also conjugated to ubiquitin, but it can also noncovalently interact with both ubiquitin family proteins. ML3 expression is induced after wounding and herbivore attack as well as by treatment with the hormone MeJA. ML3 localizes to the vacuole and to ER bodies, and ml3 mutants are defective in their response not only to herbivores but also to microbial pathogens.

ML3 Is a Novel NEDD8-and Ubiquitin-Modified Protein
We have previously used transgenic lines that express hemagglutinin-STREP-tagged NEDD8 (HSN) under the control of a dexamethasone-inducible gene expression system to isolate NEDD8-modified proteins from Arab idopsis (Aoyama and Chua, 1997;Hakenjos et al., 2011). Using the STREP affinity tag, we were able to purify a number of proteins as putative NEDD8 conjugates from total protein extracts prepared from an HSN line (Hakenjos et al., 2011). However, our experimental strategy suffered from the weakness that the choice of the purification tag only allowed purifying HSN conjugates under native conditions, so that we may have recovered proteins that are themselves not NEDD8 modified but interact with such conjugates. Therefore, we set out to examine a subset of these putative NEDD8 conjugates more critically with regard to their NEDD8 modification. In this context, we also generated MYC-ML3 lines for the expression of a MYCtagged variant of the putative NEDD8 conjugate ML3 in the HSN transgenic background. Following immunoprecipitation of the MYC-ML3 fusion protein (calculated mass of 28.3 kD), we could detect the protein with an anti-MYC antibody in two prominent forms of approximately 32 and 44 kD. Detection of HSN conjugates with an anti-hemagglutinin (HA) antibody then revealed the presence of at least two additional HSN-conjugated forms of MYC-ML3 of approximately 55 kD, suggesting that a minor fraction of MYC-ML3 is indeed NEDD8 modified (Fig. 1A). Since our further analysis of the protein sequence revealed that ML3 carries an N-terminal signal peptide that should be proteolytically cleaved during protein transport, we also generated a construct for the expression of a C-terminally tagged ML3, ML3-YFP-HA (where YFP stands for yellow fluorescent protein; calculated mass of 51 kD). Following immunoprecipitation of ML3-YFP-HA, we could again confirm that ML3 is NEDD8 modified, because a fraction of immunoprecipitated ML3-YFP-HA was also recognized by an antibody directed against the endogenous NEDD8 protein, which detected a higher mass form of ML3-YFP-HA that could be explained by the conjugation of the 8-kD NEDD8 (Fig. 1B). We thus concluded that ML3 is indeed a NEDD8-modified protein.
Following immunoprecipitation and mass spectrometry (MS), we subsequently identified Lys-137 as at least one Lys residue of immunoprecipitated ML3-YFP-HA that carried the di-Gly footprint that is retained on NEDD8-and ubiquitin-modified proteins after trypsin digestion (Supplemental Fig. S1). We then mutagenized Lys-137 and subsequently all other Lys residues of ML3 to Arg with the goal of obtaining a nonneddylatable ML3 variant. However, the NEDD8 modification of ML3 was detected in each of these ML3 mutant variants, indicating that NEDD8 may be attached to multiple or to variant Lys residues in the wild type or the mutated ML3 proteins (Supplemental Fig. S2). The conclusion that ML3 may be neddylated at multiple residues is also supported by our observation that frequently more than one neddylated form of ML3 was apparent in immunoblots following the immunoprecipitation of ML3 and the detection of HSN or endogenous NEDD8 (Fig. 1). In summary, these data suggest that ML3 may be modified by multiple NEDD8 molecules.

ML3 May Be a NEDD8-and Ubiquitin-Binding Protein
To be able to detect endogenous ML3 protein, we raised an anti-ML3 peptide antibody that readily detected the 18-kD (calculated) ML3 protein in total protein extracts from wild-type plants. Since we detected HSNconjugated ML3 following HSN immunoprecipitation, we concluded that also endogenous ML3 is NEDD8 modified ( Fig. 2A). We then used the neddylation inhibitor MLN4924 to test whether the modification of ML3 was dependent on the activity of the NEDD8 E1 conjugation enzyme (Hakenjos et al., 2011). Indeed, ML3 neddylation was reduced following a pretreatment of the plants with MLN4924, suggesting that ML3 is NEDD8 modified by the established NEDD8 conjugation pathway (del Pozo et al., 1998(del Pozo et al., , 2002del Pozo and Estelle, 1999;Woodward et al., 2007;Supplemental Fig. S3).
Interestingly, when we immunoprecipitated NEDD8 using the tagged HSN fusion variant, we noted that anti-ML3 did not only detect ML3 as a neddylated HSN conjugate but also ML3 in its apparent unmodified form ( Fig. 2A). Similar results were obtained following immunoprecipitations of HSN conjugates from the MYC-ML3 transgenic line (Supplemental Fig. S4). In these immunoprecipitations, we detected specifically the 32-kD variant of MYC-ML3 and not the 44-kD higher mass form (Supplemental Fig. S4), and we took this finding as an indirect indication that the 32-kD variant detected in MYC-ML3 lines represents the neddylatable form of MYC-ML3 (Fig. 1). The 44-kD form, in turn, may represent a posttranslationally modified form that cannot be neddylated and that could result from the protein's N-terminal signal peptide being blocked by the MYC tag.
We reasoned that the appearance of unconjugated ML3 may be the result of an enzymatic deconjugation of HSN-ML3 during the immunoprecipitation procedure. Alternatively, ML3 may bind to NEDD8 or to NEDD8 conjugates in a noncovalent manner. The latter view was supported by an EMBL-European Nucleotide Archive database entry for ML3 that described the protein as a yeast two-hybrid interactor of a ubiquitin pentamer (ABH03542) as well as by the fact that abundant amounts of ML3 were precipitated in the immunoprecipitation experiments that, in our view, could not be explained by the small amounts of NEDD8modified ML3 detected in immunoblots after HSN immunoprecipitations.
To test whether ML3 can be covalently linked to ubiquitin, we carried out immunoprecipitations from transgenic lines that express HSUB, a construct analogous to HSN for the inducible expression of an HA-STREPtagged ubiquitin. Indeed, we could detect ML3 also in the form of a high-mass ubiquitin conjugate of approximately 150 kD in the HSUB immunoprecipitates (Fig. 2B). Thus, we concluded that ML3 is a NEDD8-conjugated as well as a ubiquitin-conjugated protein in vivo. While we detected small amounts of the apparently unmodified ML3 variant also in several HSUB-ML3 purifications (data not shown), this form of ML3 was absent in others. Therefore, we reasoned that the copurification of the unmodified ML3 is the result of a deconjugation of ML3 from HSUB-ML3 (e.g. by deubiquitylating enzymes) that occurred during the immunoprecipitation of HSUB conjugates. Similarly, ML3 may be deconjugated from HSN-ML3 in the respective immunoprecipitations of HSN.
To test whether NEDD8 and ubiquitin can interact with ML3 also in a noncovalent manner, we tested their interactions using the yeast two-hybrid system. Indeed, this interaction analysis revealed that ML3 readily interacts with both ubiquitin family proteins in the yeast system (Fig. 2C). Analysis of the expression of the fusion proteins in yeast confirmed that all proteins were expressed, but interestingly, also that the GAL4 activation domain fusion protein fused to NEDD8 (AD-NEDD8) was conjugated to a broad range of proteins in the yeast host (Fig. 2D). No such conjugates could be detected in the case of activation domain-ubiquitin (AD-UBQ), possibly reflecting the fact that such ubiquitin conjugates are directly targeted for proteasomal degradation.

ML3 Belongs to the MD-2-Related Lipid-Recognition Domain Family of Proteins
ML3 is encoded by the Arabidopsis gene AT5G23820. ML3 has two closely related putative paralogs adjacent to the ML3 locus on chromosome 5 of the Arabidopsis genome, namely ML5 (AT5G23840; 76% amino acid identity to ML3) and ML6 (AT5G23830; 73% identity; Fig. 3A). Additionally, these three ML domain proteins share sequence homology with a number of proteins from Arabidopsis and B. rapa and with other proteins from non-Brassicaceae species, such as the dicots tomato (Solanum lycopersicum), grapevine (Vitis vinifera), and soybean (Glycine max) and the monocots rice (Oryza sativa) and Brachypodium distachyon ( Fig. 3; Supplemental Figs. S5 and S6). Although the genomes of both Arabidopsis and B. rapa encode three ML3-like proteins, namely Arabidopsis ML3, ML5, and ML6 and B. rapa Bra020976, Bra028884, and Bra011987, the three ML3 paralogs of each species are more closely related to each other than to the proteins from the respective other species, suggesting that they originated from independent duplication events ( Based on the current annotation of these proteins, the vast majority of ML3-related proteins from Arabidopsis and B. rapa contains an N-terminal signal peptide, which suggests that these proteins are directed to the ER and from there targeted to the vacuole or the secretory pathway ( Fig. 3A; Petersen et al., 2011). The remaining part of these ML proteins has sequence homology with the lipid recognition domain of the mammalian ML domain proteins MD-2 and NPC2 (Supplemental Fig. S6; Viriyakosol et al., 2001;Inohara and Nuñez, 2002). However, the restricted homology within this domain does not allow drawing any conclusions on the binding specificity of any of the Arabidopsis proteins (Supplemental Fig. S6). Based on their homology to MD-2, the members of this protein family were previously classified as ML proteins; however, none of the plant ML domain proteins have been characterized at the biochemical, cell biological, or functional level as yet (Inohara and Nuñez, 2002). We also noted with interest that proteins closely related to ML3 can only be found in Brassicales species, including Arabidopsis and B. rapa, or inversely that these proteins are seemingly absent from the genomes of non-Brassicales dicots, monocots, the moss Physcomitrella patens, and the lycophyte Selaginella moellendorffii ( Fig. 3B; Supplemental Fig. S5).

Identification of ml3 Mutants
In order to understand the biological function of ML3, we examined two putative ml3 alleles with transfer DNA (T-DNA) insertions in the ML3 open reading frame and 39 untranslated region, ml3-3 (SALK_001255) and ml3-4 (SAIL_182_G07; Fig. 4A). Our subsequent analysis of ML3 protein abundance in the ml3-3 and ml3-4 mutants showed that ML3 protein cannot be detected in ml3-3 and ml3-4 ( Fig. 4B). While this work was in progress, two additional mutant alleles, ml3-1 (SALK_059591) and ml3-2 (SALK_091638), were described as part of another study of genes related to the defense to herbivores (Fridborg et al., 2013). While this study had associated semidwarfed growth with ml3-1 and ml3-2, we could not detect such a phenotype in ml3-3 and ml3-4 when grown in continuous light growth conditions (Fig. 4C). For our further work, we considered the exon insertion allele ml3-3 as the prototypical ml3 mutant allele.

ML3 Resides in the Vacuole and in ER Bodies
In order to understand the intracellular distribution of ML3, we generated transgenic lines that express ML3 as a fusion with the fluorescent protein mCherry under the control of a 2-kb ML3 promoter fragment (ML3p:ML3-mCherry). Our analysis of transgenic lines expressing this protein revealed that the protein accumulates in the vacuole as well as in rod-shaped structures that we identified as ER bodies based on their size and shape. This was also supported by their partial colocalization with the marker Q4, which highlights the membranes of ER bodies as well as the cellular ER network, and by their exclusive presence in the seedling epidermis and the cotyledons (Fig. 5A). The ER body localization of ML3 was also verified in lines expressing ML3-YFP from the ML3 promoter fragment (ML3p:ML3-YFP; Fig. 5B). In turn, ML3p:ML3-YFP lines did not allow us to visualize the vacuolar localization of ML3 (Fig. 5C), most likely due to the fact that GFP and the closely related YFP are degraded in the vacuole (Tamura et al., 2003). To confirm the vacuole and ER body localization for the endogenous ML3, we purified vacuoles and performed a subcellular fractionation based on previously established protocols Robert et al., 2007). In these analyses, we identified endogenous ML3 in vacuolar preparations from 14-d-old seedlings where ML3 cofractionated with the marker vacuolar ATPase (Fig. 5D). To resolve ER bodies, we performed centrifugation of protein extracts at 1,000g from a transgenic line expressing A, ClustalOmega alignment of ML3 (AT5G23820) and its homologous sequences retrieved from Arabidopsis (AT) and B. rapa (Bra). The N-terminal signaling peptide (shaded) was predicted with SignalP 4.0 (Petersen et al., 2011). The seven Lys (K) residues that are conserved in ML3, ML5, and ML6 but also in other MD-2 domain proteins are indicated. Cys residues required for the formation of two intramolecular Cys bridges that have been reported for MD-2 also appear to be conserved in the plant proteins. An exception are the Arabidopsis proteins ML3, ML5, and ML6 that lack one of these conserved Cys residues; Cys-100 (ML3) may functionally replace this missing Cys residue. B, Phylogenetic tree of the MD-2 domain proteins related to ML3 from Arabidopsis (AT) and B. rapa (Bra) as shown in A as well as from the moss P. patens (PP), soybean (GLYMA), rice (Os), B. distachyon (BRADI), grapevine (Vv), and tomato (Solyc). Protein sequences were retrieved at www.ensembl.org. The phylogenetic tree was generated with the conserved MD-2 domain and is drawn to scale. The underlying alignment is shown in Supplemental Figure S5. Bootstrap values are indicated by each node. Bar = 0.2 amino acid substitutions per site.
the ER marker GFP-HDEL. Interestingly, ML3 could not be retrieved in the 1,000g fraction that contains the ER bodies, suggesting that native ML3 may not reside in ER bodies and that the ER body localization of ML3-mCherry may be an artifact of the protein fusion (Fig. 5D). In turn, we detected ML3 in the soluble S100 fraction, in line with its localization to the vacuole. Identical fractionations of the YFP-tagged ML3-YFP-HA, however, confirmed the proposed ER body localization also for ML3-YFP-HA (Fig. 5E). In summary, we concluded that ML3 is a vacuolar protein that may potentially also reside in ER bodies.
The vacuolar localization of ML3 observed here is in agreement with the results of a previously published proteomics study that had identified ML3 in vacuoles of Arabidopsis plants (Carter et al., 2004). Interestingly, ML3 was also identified in another proteomics study that aimed at the identification of proteins that are secreted in response to treatments with the plant hormone salicylic acid (SA; Oh et al., 2005). Since ML3 had been shown to be involved in the plant's defense to herbivore attack, which induces SA responses (Oh et al., 2005;Bejai et al., 2012;Fridborg et al., 2013), we reasoned that ML3 may relocalize to the extracellular space in response to activation of the SA pathway. While we noticed indeed a presence of ML3-mCherry after SA treatment (2 h), our subsequent analyses indicated that this is likely artifactual, because it was also found with the unrelated vacuolar lumen marker sp-RFP-AFVY (Hunter et al., 2007) and because ML3-mCherry secretion was the apparent result of pH changes and did not occur in buffered solutions or in the presence of mannitol (Supplemental Fig. S7).

ML3 Is Coexpressed with ER Body Genes and in Response to NAI1
To gain further insight into the function of ML3, we searched for genes that are coexpressed with ML3. Through the analysis of all available microarray data sets deposited in Genevestigator, we identified a number of coexpressed genes and noticed with interest that eight genes among the first 20 coregulated genes had a demonstrated or proposed function in ER body biology or formation (Table I). This set of coregulated genes included NAI1, a basic helix-loop-helix transcription factor essential for ER body formation, as well as proposed NAI1 target genes such as NAI2 and PYK10 (Matsushima et al., 2004). Since ML3 had also been found to be differentially expressed in a comparison of nai1 mutants with the wild type, we hypothesized that NAI1 may regulate ML3 gene expression. This assumption found support in our observation that ML3 expression was strongly reduced in the nai1 mutant nai1-3 (GK-136G06-012754; Fig. 6A). Since nai1 mutants are defective in the formation of ER bodies and since we had identified ML3 The merged image is shown in the right panel. The double arrowheads point at ER bodies. B, Confocal image of the expression of ML3p:ML3-YFP in the epidermis of a 5-d-old Arabidopsis seedling. C, Immunoblots of total protein extracts and a vacuole preparation from 14-d-old Arabidopsis wild-type seedlings. D and E, Immunoblots of a total protein extract and pellet fractions obtained after differential centrifugation of a protein extract prepared from 7-d-old GFP-HDEL (D) and ML3-YFP-HA (E) transgenic seedlings after centrifugation at 1,000g (P1), 8,000g (P8), or 100,000g (P100). S100, Soluble supernatant after centrifugation at 100,000g; T, total protein extract. as a putative ER body protein, we questioned whether the apparent absence of ML3 protein in the nai1 mutant was the consequence of the absence of NAI1 as its transcriptional regulator or the indirect consequence of the absence of ER bodies in this mutant. To address this question, we examined ML3 protein abundance also in nai2-2 (SALK_005896) and nai2-3 (SALK_043149) mutants. NAI2 is a protein of unknown function in ER bodies that is essential for ER body formation but does not have an apparent function related to transcriptional regulation (Yamada et al., 2008). Importantly, ML3 protein was detectable in nai2 but, as mentioned above, absent in nai1, suggesting that the absence of ML3 in nai1 is the result of the absence of the transcriptional regulation by NAI1 rather than the indirect consequence of the absence of ER bodies (Fig. 6C). Inversely, we found no evidence that the absence of ML3 in ml3-3 has an influence on the formation of ER bodies (Supplemental Fig. S8).
NAI1 is essential for ER body formation, which is promoted by the plant hormone jasmonic acid (JA; Matsushima et al., 2004). It has previously been established that NAI1 transcription is induced by MeJA and reduced in response to ethylene (Matsushima et al., 2004). We hypothesized that ML3 expression should follow this pattern of transcriptional regulation of NAI1 if it was a direct NAI1 transcription target. Indeed, we found that the abundance of ML3 and NAI1 as well as the ER body-resident PYK10 gene is induced by MeJA treatments and suppressed by concomitant treatments with the ethylene precursor 1-aminocyclopropane-1-carboxylic acid, which is converted into the gaseous ethylene hormone when taken up by the plant (Fig. 7A). This finding thus supported the notion that ML3 is a direct NAI1 transcription target.
MeJA is produced in plants in response to wounding by herbivore pathogens. The knowledge about the regulation of NAI1 and ER body formation by MeJA has given rise to the concept that ER bodies may play a role in the response to pathogens (Yamada et al., 2011). Interestingly, we found ML3 expression in the ML3p: ML3-mCherry lines to be restricted to the epidermis, an observation that could be considered in support of the proposed function in biotic and abiotic interactions with the environment (Fig. 7B). To assess the possibility of a putative regulation of ML3 expression by wounding, we generated transgenic lines that express the reporter GUS under the control of a 2-kb ML3 promoter fragment. Following the wounding of leaves of transgenic lines expressing this ML3p:GUS construct, we observed indeed increased staining at the sites of wounding as well as systemic spreading of the signal over the entire blade of the wounded leaf (Fig. 7C). Since ML3 had previously been analyzed in the context of herbivore defense, we also challenged ML3p:GUS lines with second instar larvae of the herbivore Spodoptera littoralis and found clear evidence for ML3 induction at the sites of wounding after insect feeding (Fig. 7D). In summary, these findings support that notion that ML3 is a transcription target downstream from NAI1 that is induced in response to wounding and herbivore attack.
A Role for ML3 in the Pathogen Response ER bodies have been proposed to function in pathogen responses (Yamada et al., 2011). For this reason, we were interested in examining the performance of ml3 mutants after challenge with the fungal pathogen Alternaria brassicicola and the pathogenic bacterium Pseudomonas syringae DC3000. A. brassicicola is a necrotrophic fungus that induces cell death upon infection. In comparison with the wild type, in which cell lesions were largely restricted to the inoculation spots, ml3 mutants showed increased susceptibility, visible by the increased spreading of necrosis away from the site of infection at 5 d after infection (Fig. 8, A and B). Lesion expansion was associated with an increase of fungal DNA in ml3 mutants and was partially similar to the responses of the camalexindeficient mutant pad3-1 that has previously been shown to possess enhanced susceptibility to A. brassicicola (Schuhegger et al., 2006). In the case of the P. syringae infection experiments, we detected reduced pathogen growth 6 d post infiltration in the ml3 mutants when compared with the wild type (Fig. 8, C and D). In quantitative terms, this reduced growth was comparable to the reduced growth observed in the coi1-1 mutant, an established mutant of the P. syringae pathogen response (Xie et al., 1998). Based on these observations, we concluded that ML3 has a role in pathogen response in Arabidopsis.

DISCUSSION
In this study, we identify ML3 as a NEDD8-and ubiquitin-modified protein that can be detected in the vacuole and in ER bodies of epidermal cells in Arabidopsis seedlings. ML3 is related to the mammalian proteins MD-2 and NPC2 ( Fig. 3; Supplemental Fig. S6). MD-2 recognizes bacterial lipopolysaccharides together with TLR4 as part of the mammalian innate immune response (Viriyakosol et al., 2001;Kim et al., 2007). NPC2 binds to cholesterol in the mammalian lysosome and participates in intracellular cholesterol transport together with the transport protein NPC1 (Frolov et al., 2003;Xu et al., 2007). Thus, ML domain proteins bind to different ligands and localize to diverse cellular compartments. Many ML domain proteins can be Figure 6. ML3 expression is promoted by the ER body regulator NAI1. A and B, Results of qRT-PCR analyses testing for NAI1 (A) and ML3 (B) transcript abundance in 7-d-old wild-type (wt) and nai1-3 or ml3-3 mutant seedlings. C, Immunoblot with anti-ML3 testing for the abundance of ML3 in total protein extracts (45 mg) prepared from 7-d-old wild-type seedlings as well as nai1 and nai2 mutant seedlings. Anti-CDC2 served as a loading control for this experiment. identified in the genomes of a variety of plants (Fig. 3), but none of these proteins have been studied at the biochemical or cell biological level to date. Although their overall homology to the better studied mammalian proteins allows suggesting that their biochemical function as binding proteins for a hydrophobic ligand may be conserved, the restricted degree of homology does not allow drawing conclusions on their putative binding partner or transport cargo (Supplemental Fig. S6). Interesting, and possibly also helpful for the identification of candidate ligands for ML3, is the fact that ML3 and its closest homologs appear to form a Brassicales-specific subfamily of ML domain proteins within the larger plant ML protein family ( Fig. 3; Supplemental Fig. S5). Thus, also, the ligand of ML3 may be specific for the order Brassicales.
Our cell biological analyses revealed that ML3 is a vacuolar protein that we detected using mCherry fusions of the protein. The vacuolar localization of ML3 could also be confirmed for the endogenous protein in biochemical fractionation experiments. The localization of ML3 to the vacuole is in agreement with the presence of an N-terminal signal peptide that suggests targeting of ML3 to the ER, from where it may be targeted to the vacuole or the secretory pathway (Petersen et al., 2011). This signal peptide is also conserved in the vast majority of ML domain proteins from plants (Fig. 3). The vacuolar localization of ML3 is also in line with the finding that ML3 had previously been reported by others to be a component of the vacuolar proteome (Carter et al., 2004). Thus, similar to mammalian NPC2, ML3 resides in the lysosomal compartment of the cell, where it may retain a specific ligand or participate in a transport process.
We also observed ML3-mCherry and ML3-YFP in ER bodies that were recognizable based on their size and shape in epidermal cells of Arabidopsis seedlings (Fig. 5). Our attempts to confirm the ER body localization for the endogenous ML3 protein by biochemical methods were not successful. This inability to detect endogenous ML3 in ER body fractions using biochemical methods may reflect the fact that ER bodies are only present in epidermis cells and, additionally, represent only a minor fraction of the cellular proteome. Thus, the abundance of ML3 in ER bodies, particularly in comparison with the amount of vacuolar ML3, may be too small to be detected after fractionation on immunoblots with the ML3-specific antibody. For a number of reasons, we consider our observation of the ER body localization of the fluorescent protein-tagged ML3 variants biologically significant. First, ML3 expression is strongly coregulated with a number of genes that have a known and, in part, demonstrated function in ER body biology, including NAI1 and NAI2, two proteins required for ER body formation (Table I). NAI1 has been proposed to regulate the expression of a number of ER body genes, and ML3 was previously identified as one of the most strongly down-regulated genes in a gene expression analysis of nai1 mutants . In line with this finding, ML3 gene expression and protein abundance are strongly reduced in nai1 mutants, and ML3 expression follows the dynamics of NAI1 regulation by MeJA and ethylene (Figs. 6 and 7). ML3 expression is also coregulated with that of NAI2, a protein of unknown biochemical function that is also required for ER body formation (Yamada et al., 2008). Although nai2 mutants lack ER bodies, ML3 protein is still detectable in this mutant, indicating that ML3 abundance is not dependent on the presence of ER bodies and that the absence of ML3 in nai1 is most likely the result of the absence of NAI1 as a dominant transcriptional regulator of ML3. In turn, the presence of ER bodies is not affected by the presence or absence of ML3, as revealed by the presence of GFP-HDEL-positive ER bodies in the ml3-3 mutant (Supplemental Fig. S8). ML3 expression is furthermore coregulated with the b-glucosidase PYK10/ BLU23, an ER body-specific glucosidase (Sherameti A, Representative photographs of leaves from 3-week-old Arabidopsis plants 5 d post infection with 10 mL of a 5 3 10 5 spores mL 21 A. brassicicola suspension. B, qRT-PCR analysis for the A. brassicicola genomic DNA from 100 mg of plant material as a quantitative measure for fungal growth on infected leaves. C, Representative photographs of leaves from 3-week-old Arabidopsis plants 6 d post infiltration with 10 7 colony-forming units mL 21 P. syringae DC3000 suspension. D, qRT-PCR analysis of P. syringae DC3000 genomic DNA from 100 mg of plant material as a quantitative measure for bacterial growth on infiltrated leaves. dpi, Days post infection; wt, wild type. et al., 2008), the gene encoding MEMBRANE OF ER BODY1 (Yamada et al., 2013), as well as several Jacalin-related lectins and GDSL lipase, which all have been associated with the biology of ER bodies (Nagano et al., 2005Yamada et al., 2011). In summary, we conclude that ML3 has a function in ER body biology.
ER bodies have a proposed role in pathogen responses (Yamada et al., 2011). In line with this, we could reveal altered pathogen responses in ml3 mutants after treatment with the pathogens A. brassicicola and P. syringae DC3000. Plants often employ distinct recognition mechanisms and signaling pathways for different pathogen elicitors. In this context, the plant hormones SA, JA, or JA/ethylene appear to form a network of synergistic and antagonistic interactions (Glazebrook, 2001;Spoel et al., 2003;Spoel and Dong, 2008;Leon-Reyes et al., 2009). Plants restrict the colonization of some pathogens like P. syringae through the elicitation of SA defense responses (Shirasu et al., 1997). Whereas JA-mediated defenses are necessary to inhibit fungal pathogens like A. brassicicola (van Wees et al., 2003), in this study, we provide evidence that ml3-3 mutants are compromised in their JA-mediated defense responses by their enhanced susceptibility toward A. brassicicola (Fig. 8A). P. syringae is known to suppress SA responses in plants by inducing coronatine-mediated JA signaling (Fig. 8B). The resistant phenotype seen in ml3 mutants upon P. syringae infection is congruent with other studies showing the antagonistic effects of JA and SA signaling (Mur et al., 2006;Spoel et al., 2007;Koornneef et al., 2008).
Interestingly, our results also show that a number of genes of the glucosinolate pathway are strongly coregulated with ML3, including the genes for the nitrile specifier proteins NSP1, NSP3, and NSP4 that convert glucosinolates to nitrile (Kuchernig et al., 2012), (Pfalz et al., 2011), the glucosinolate biosynthesis regulatory transcription factor ATR/MYB34 (score 0.564; Bender and Fink, 1998;Celenza et al., 2005), as well as the ER body marker PYK10 (Sherameti et al., 2008;Table I). Glucosinolate hydrolysis at pH , 5 leads to the formation of nitriles by the activation of the nitrile specifier proteins (Halkier and Gershenzon, 2006;Kissen and Bones, 2009), and in this respect, our observation that a pH shift to 3.5 can lead to changes in the distribution of ML3-mCherry may be biologically significant (Supplemental Fig. S7). PYK10 protein has previously been shown to possess enhanced activity upon hyperinfection by symbiotic fungi by activating toxic compounds such as indole glucosinolates (Ahn et al., 2007;Sherameti et al., 2008). Other studies have revealed that camalexin-and indole glucosinolate-deficient plants are hypersusceptible to the fungus Sclerotinia sclerotiorum (Stotz et al., 2011). In addition, JA signaling is essential for the induction of indole glucosinolates (van Dam and Oomen, 2008). Previous studies have shown that ml3 mutants are compromised in woundinduced JA signaling and are more appetizing for the generalist larva S. littoralis (Fridborg et al., 2013). However, plants compromised in ML3 expression did not show any antiherbivory traits toward the specialist herbivore Plutella xylostella (Fridborg et al., 2013). This could be attributed to the fact that Plutella spp. have evolved to combat the glucosinolate-myrosinase defense (Kliebenstein et al., 2002). Based on these studies and our observations, we speculate that ML3 is a positive regulator downstream from JA responses and glucosinolate-mediated plant defenses. This suggested link between ML3 and the biology of glucosinolates may also be helpful for the identification of a putative ML3 ligand, which may be a glucosinolate or a glucosinolate-related metabolite, compounds that are known to be stored in the vacuole (Grubb and Abel, 2006).
Our interest in ML3 was triggered following our identification of the protein as a putatively NEDD8modified protein. ML3 had also been found by others in similar attempts to identify novel NEDD8 conjugates (Hotton et al., 2012). Our experiments now clearly identify a small fraction of ML3 as NEDD8 modified (Fig. 1). Interestingly, we identify also a ubiquitinconjugated form of ML3 that does not appear to be neddylated, and we are able to show that ML3 can interact with NEDD8 as well as with ubiquitin in the yeast two-hybrid system. Since yeast also contains active neddylation as well as ubiquitylation systems, these interactions may also be the result of conjugation between the respective bait and prey fusion proteins. The modification and binding of both ubiquitin-related proteins to ML3 may be suggestive for a regulatory interplay between these modifier proteins. Unfortunately, we were unable to express and purify ML3 using recombinant expression systems, and this has prevented us from further investigating the interactions between ML3 and the ubiquitin family proteins. Noncovalent interactions of ML3 with NEDD8 and, therefore, an anticipated enrichment of ML3 in purifications of NEDD8 conjugates may be the reason for the identification of ML3 in both proteomics studies aiming at the identification of novel NEDD8 targets from plants (Hakenjos et al., 2011;Hotton et al., 2012). We note that also the ML3-related ML6 was identified unequivocally in our initial MS analysis following HSN purification and that, thus, the findings reported here for ML3 may also apply to the paralogs ML6 and ML5 (Hakenjos et al., 2011;Hotton et al., 2012). Since ML3 does not have any protein features that would be indicative for a binding activity to ubiquitin or ubiquitin-like proteins, these observations as well as the fact that the protein is NEDD8 and ubiquitin modified are very intriguing. At the same time, they give rise to a number of questions about the role and the possible interplay of these modifications and interactions for ML3 biology and function. Future research will have to address these important and exciting issues.

Cloning Procedures
To generate MYC-ML3, an ML3 entry clone (G13160) was obtained from the Arabidopsis Biological Resource Center and then cloned into pJawohl2B-5xMYC-GW using Gateway technology ( . ML3-YFP-HA was obtained by insertion of a PCR fragment obtained with primers 11 and 12 into the Gateway-compatible vector pEarleyGate101 (Earley et al., 2006). The constructs for the expression of the ML3 promoter-driven ML3-YFP (ML3p:ML3-YFP) and ML3-mCherry (ML3p:ML3-mCherry) were generated in the following manner. An ML3 promoter gene fragment was amplified by PCR from Arabidopsis ecotype Columbia genomic DNA with the primers 33 and 35 that added a HindIII and a NheI restriction site to the promoter gene termini, respectively. A PCR fragment of the mCherry or YFP coding sequence flanked by NheI and BamHI restriction sites was obtained with primers 38 and 39 (mCherry) or 40 and 41 (YFP), and a 1-kb ML3 terminator fragment flanked by BamHI and XhoI sites was amplified using primers 36 and 37. All fragments were ligated into the HindIII and XhoI restriction sites of pGreen0029 (mCherry) or pGreen0229 (YFP), and the final construct was transformed into the Q4 ER marker line (ML3p:ML3-mCherry; Cutler et al., 2000) or ml3-3 (ML3p:ML3-YFP; Hellens et al., 2000).
ML3p:GUS was generated by inserting a 2-kb promoter fragment amplified by PCR (primers 33 and 34) from Arabidopsis Columbia genomic DNA into the vector pCAMBIA1391Z (www.cambia.org/daisy/cambia/). At least 10 transgenic Arabidopsis plants were generated for each construct using the floral dip transformation protocol. T1 seeds were selected for resistance to the respective antibiotic or herbicide as well as for transgene expression. Individual lines were chosen for cell biological and biochemical analyses and for genetic crosses. Primer sequences for cloning are listed in Supplemental Table S1.
The yeast two-hybrid construct BD-ML3 was obtained by PCR amplification of the ML3 open reading frame with oligonucleotides 13 and 14 and was then inserted into pGBKT7 as an EcoRI and XhoI fragment. AD-NEDD8 and AD-UBQ were cloned in a similar manner by ligation of the NEDD8/RUB1 (AT1G31340) and UBQ (AT3G52590) open reading frames as EcoRI-XhoI/SalI fragments into pGADT7 AD. Primer sequences are listed in Supplemental Table S1.

Quantitative Real-Time PCR
Extraction of total RNA and complementary DNA synthesis were conducted as described previously (Richter et al., 2010). The complementary DNA equivalent of 25 ng of total RNA was used in a 10-mL PCR in a CFX96 Real-Time System Cycler with iQ SYBR Green Supermix (Bio-Rad). A 40-cycle two-step amplification protocol (10 s at 95°C, 30 s at 60°C) was used for all measurements. Primer sequences are listed in Supplemental Table S1. Unless otherwise stated, the average and SE of four technical replicates pooled from at least two biological replicates are shown. The experiment was repeated at least once, and the result of a representative experiment is shown.
Subcellular fractionation from 7-d-old seedlings was performed as described previously . Vacuoles were purified from 12-to 14-dold seedlings using a Ficoll gradient as described previously, and vacuolar proteins were subsequently precipitated using TCA (Robert et al., 2007).

MS
For protein digestion, a sample of immunoprecipitated ML3-YFP-HA was reduced and alkylated by 50 mM dithiothreitol and 10 mg mL 2 1 chloroacetamide, respectively. Tryptic in-gel digestion was performed according to standard procedures. Nanoflow liquid chromatography-tandem MS was performed using an Eksigent nanoLC-Ultra 1D+ system coupled online to an LTQ-Orbitrap Velos (Thermo Scientific) mass spectrometer. Tryptic peptides were dissolved in 20 mL of buffer A (0.1% formic acid in double-distilled water), and 10 mL was injected for each measurement. Peptide samples were first loaded on a trap column (100 mm i.d. 3 2 cm, packed in house with 5 mm Reprosil PUR AQ; Dr. Maisch) in 100% buffer A. Peptides were transferred to an analytical column (75-mm 3 40-cm C18 column, 3 mm Reprosil PUR AQ Gold; Dr. Maisch) and separated using a 225-min gradient from 7% to 35% buffer B (0.1% formic acid in acetonitrile). MS measurements were performed in data-dependent acquisition mode, automatically subjecting the 10 most abundant precursor ions in the full MS spectra for higher-energy collisional dissociation fragmentation at 30% collision energy. Full MS spectra and tandem MS spectra were acquired at 30,000 and 7,500 resolution, respectively. Intensitybased label-free quantification was performed using Progenesis (version 4.0; Nonlinear Dynamics). The generated peak list was then searched using Mascot (version 2.4.1) against the protein sequence databases National Center for Biotechnology Information non-redundant (download October 26, 2011; 15.8 million sequences) and SwissProt (version 57; 0.5 million sequences) for protein identification. The variable modification of Lys(GlyGly) was considered in the database search in order to identify NEDD8-modified ML3.

Phylogenetic Analysis
Protein sequences containing the MD-2-related lipid-recognition domain (PF02221) were retrieved from the Ensembl BioMart database for the dicot species Arabidopsis, Brassica rapa, soybean (Glycine max), tomato (Solanum lycopersicum), and grapevine (Vitis vinifera), the monocot species Brachypodium distachyon and rice (Oryza sativa), as well as the moss Physcomitrella patens and the lycophyte Selaginella moellendorffii. Sequences were aligned with ClustalOmega (www.ebi.ac.uk/Tools/msa/clustalo/) and manually amended when necessary. The phylogenetic tree was constructed based on the amended alignment with MEGA5 using the maximum likelihood method and the bootstrap method with 1,000 bootstrap replications. The tree was rooted with the sequences from P. patens and S. moellendorffii. The cutoff values for the bootstrap analysis were set to 60%, and bootstrap values are indicated by each node.

Genevestigator
A list of 20 genes coexpressed with ML3 was generated on the basis of 8,689 Affymetrix 22k ATH1 microarray samples using the Genevestigator coexpression tool (www.genevestigator.com).

Chemical Treatments
To examine the effects of MeJA and ethylene on ML3 expression, rosette leaves of 16-d-old plants were floated for 12 h on water containing 50 mM MeJA. The ethylene precursor 1-aminocyclopropane-1-carboxylic acid (50 mM) was added for the combined treatments with MeJA and ethylene. To examine the cellular distribution of ML3-mCherry and RFP-AVFY, 5-d-old seedlings were treated for 2 h with SA (500 mM), 3-hydroxybenzoic acid (500 mM) dissolved in water, germination medium (4.3 g L 21 Murashige and Skoog medium, 10 g L 21 saccharose, and 0.5 g L 21 MES, pH 5.8) or 0.5 M mannitol or, alternatively, in water, pH 3.5 (HCl). Dexamethasone-inducible transgenes were induced by immersing the plant material overnight in liquid germination medium supplemented with 30 mM dexamethasone (Sigma; Aoyama and Chua, 1997).

Yeast Two-Hybrid Assay
The yeast two-hybrid assay was performed as described previously .

Pathogen Assays
Alternaria brassicicola strain MUCL20297 was grown on potato dextrose agar plates for 2 weeks at 22°C. Spores were then harvested and suspended in water (5 3 10 5 spores mL 21 ). A. brassicicola inoculation was performed by adding 5-mL drops onto the leaf surface as described previously (Thomma et al., 1998). Two leaves per plant were inoculated with two drops each. Ten plants were taken as one biological replicate, and three biological replicates were used. Five days after challenge, disease severity was scored and samples were collected for pathogen quantification. Disease rating was assessed on the basis of symptom severity ( Van der Ent et al., 2008). For each pathogen assay, inoculated leaves of 10 plants were pooled together and taken as one biological replicate, and three biological replicates were used. Fungal DNA quantification was carried out by quantitative real-time (qRT)-PCR of ABU03393 using the primers 48 and 49. In preparation for a Pseudomonas syringae pv tomato DC3000 experiment, the bacteria were grown overnight in Luria-Bertani medium at 37°C. The cells were centrifuged, and the pellet was washed once with sterile 10 mM MgCl 2 and resuspended in 10 mM MgCl 2 to an optical density at 600 nm of 0.02 (10 7 colony-forming units mL 21 ). Inoculations for the P. syringae pv tomato DC3000 bioassays were performed through pressure infiltration with bacterial suspension into the tissue of the abaxial part of the rosette leaves (two leaves per plant). About 100 mL of suspension was used for each leaf, and 10 plants were infiltrated for one biological replicate. Three biological replicates were used in the study. The experiment was carried out twice, and a representative experiment is shown. Pathogen quantification for P. syringae pv tomato DC3000 was performed by qRT-PCR of NC_004578 using the primers 50 and 51.

Supplemental Data
The following materials are available in the online version of this article.
Supplemental Figure S5. Protein alignment of the ML domain used for the phylogeny.
Supplemental Figure S6. Protein sequence alignment of ML3 with human and Arabidopsis ML domain proteins.
Supplemental Figure S7. Acidic pH causes relocalization of vacuolar proteins.
Supplemental Table S1. List of primers.