Arabidopsis PIAL1 and 2 Promote SUMO Chain Formation as E4-Type SUMO Ligases and Are Involved in Stress Responses and Sulfur Metabolism[C][W][OPEN]

Substrates modified by covalent attachment of the small ubiquitin-related modifier (SUMO) usually receive a single SUMO moiety. Existence of enzymes that promote formation of SUMO chains indicates that such chains have distinct roles in plants, and mutant analysis indicates involvement in regulation of sulfur metabolism and stress responses. The Arabidopsis thaliana genes PROTEIN INHIBITOR OF ACTIVATED STAT LIKE1 (PIAL1) and PIAL2 encode proteins with SP-RING domains, which occur in many ligases of the small ubiquitin-related modifier (SUMO) conjugation pathway. We show that PIAL1 and PIAL2 function as SUMO ligases capable of SUMO chain formation and require the SUMO-modified SUMO-conjugating enzyme SCE1 for optimal activity. Mutant analysis indicates a role for PIAL1 and 2 in salt stress and osmotic stress responses, whereas under standard conditions, the mutants show close to normal growth. Mutations in PIAL1 and 2 also lead to altered sulfur metabolism. We propose that, together with SUMO chain binding ubiquitin ligases, these enzymes establish a pathway for proteolytic removal of sumoylation substrates.

The spectrum of identified sumoylation substrates (Budhiraja et al., 2009;Elrouby and Coupland, 2010;López-Torrejón et al., 2013;Miller et al., 2010Miller et al., , 2013 suggests that a wide variety of proteins can be modified by SUMO, with emphasis on proteins of the nucleus. Substrates typically carry a single SUMO moiety. However, the recent discovery of ubiquitin ligases that bind to SUMO chains (Yin et al., 2012b;Elrouby et al., 2013;Sriramachandran and Dohmen, 2014) suggests that SUMO chains have physiological roles as well. However, it is currently unclear how widespread SUMO chains really are, how their formation is regulated, and how they influence biological processes in plants.
In contrast to the analogous ubiquitylation process, which employs a plethora of E3 ligase components to target specific substrates, sumoylation relies on a surprisingly small number of components, contrasting with its broad spectrum of substrates. In particular, SCE, the product of an essential single copy gene in Arabidopsis thaliana (Saracco et al., 2007), can directly transfer SUMO to a significant fraction of pathway substrates. To date, only two E3 ligases of the SUMO pathway have been described in Arabidopsis, SAP AND MIZ1 (SIZ1) (Miura et al., 2005) and HIGH PLOIDY2 (HPY2) (Ishida et al., 2009;called MMS21 in Huang et al., 2009). Both proteins contain an SP(SIZ-PIAS)-RING (also called a zf-MIZ) domain, which groups them together with mammalian PIAS (protein inhibitor of activated STAT) type SUMO ligases. Whereas HPY2 regulates endoreduplication cycles and affects root development, SIZ1 is best known for involvement in a variety of stress responses (see above).
The Arabidopsis genome encodes at least two additional proteins with SP-RING domains; we called these proteins PROTEIN INHIBITOR OF ACTIVATED STAT (PIAS) LIKE1 (PIAL1) and 2 and examined their contribution to sumoylation in plants. We find that both function in vitro to enhance the formation of SUMO chains, i.e., SUMO-SUMO isopeptide linkages. Mutant analysis suggests that PIAL1 and 2 contribute to the regulation of salt stress and osmostress responses and are involved in sulfate assimilation and sulfur metabolism. Their functions overlap, but show no obvious redundancy with the previously characterized SUMO ligase SIZ1, which is consistent with the model that SIZ1 preferentially enhances substrate sumoylation, whereas PIALs extend SUMO residues on substrates into chains.

Two Arabidopsis Genes Encoding Putative SUMO Ligases
The Arabidopsis genome encodes four proteins with an SP-RING (zf-MIZ) domain. Structural studies indicate that this domain mediates binding to the SCE (Yunus and Lima, 2009) to promote SUMO conjugation. Two of the Arabidopsis SP-RING-containing proteins were characterized in previous work; to complement this, we studied the physiology and biochemistry of the other two proteins. At1g08910 was called PIAL1, and At5g41580 was called PIAL2. Their mRNAs were identified via cDNA isolation, showing that the two putative genes are not pseudogenes. The PIAL2 cDNA was obtained from the RIKEN repository (Seki et al., 2002) and the PIAL1 cDNA was isolated by reverse transcription of RNA prepared from plants. Sequences of their open reading frames, as obtained from the cDNAs, and protein sequences are shown in Supplemental Figures 1 and 2. The two proteins have similarity to each other (Figure 1; Supplemental Figure 3A). One striking difference between them is the presence of an insertion consisting of seven repeats of 23 amino acids in PIAL1, but not in PIAL2 (Supplemental Figure 3C). PIAL1 and 2 have potential orthologs in other plants, although some plants have only a single gene, as opposed to the two representatives in Arabidopsis (Novatchkova et al., 2012). Using the SP-RING sequence for alignment, Figure  1C shows a phylogenetic tree, indicating that plants encode three classes of SP-RING containing SUMO ligases. Branch lengths reflect the number of substitutions (scale bar indicates 0.2% changes per base). Entries from Arabidopsis (At), tomato (Solanum lycopersicum; Solyc), rice (Oryza sativa; Os), and sorghum (Sorghum bicolor; Sb) were used for tree building. Bootstrap values are listed in Supplemental Figure 4, and gene candidates from additional species can be found in Novatchkova et al. (2012). The PIAL1 and 2 cDNA sequences differ slightly from previously annotated sequences or gene models (see Supplemental Figure 5 for comparative alignment of proposed PIAL1 intron-exon structures). However, we show in the following that PIAL1 and PIAL2 proteins are functional SUMO ligases in vitro, indicating that the sequences characterized in this work represent authentic gene products. Both genes are expressed at low to moderate levels in a broad range of tissues including young rosette leaves, cauline leaves, and flowers (Supplemental Figure 6). The Univ. of Toronto eFP Browser (Winter et al., 2007) indicates that transcription of PIAL1 is increased by heat, whereas transcription of PIAL2 is transiently increased by salt and osmotic stress.
T-DNA insertions in both genes were obtained from public collections (Sessions et al., 2002;Alonso et al., 2003;Kleinboelting et al., 2012), and we used two independent alleles for assays of this (B) RT-PCR experiment (three technical replicates) to demonstrate absence of full-length mRNA in mutants pial1-1, pial1-3, pial2-1, and pial2-2. M, DNA size marker (bands of 3, 2, 1.5, and 1 kb). work in order to exclude the influence of unintended mutations that might be present in the studied plant lines. Positions of T-DNA inserts are shown as thick arrows in Figure 1A. None of these mutant lines expresses full-length mRNA ( Figure 1B; three technical replicates shown). Neither pial1 or pial2 single mutants nor double mutants showed any obvious growth difference under standard greenhouse conditions (Supplemental Figure 7). The slightly reduced growth of pial1 pial2 double mutants on plates at early stages ( Figure  2A) does not translate into differences of mature plants on soil. This contrasts with significant growth defects of the previously identified siz1 mutant. In the siz1 background, additional mutations in pial1 and pial2 lead to a further, albeit small, reduction in size. All mutant combinations with siz1 are viable (Supplemental Figure 7) and set seed if supplemented with ammonium ions (Park et al., 2011).

Phenotypic Characteristics of pial Mutants
Based on the known role of SUMO conjugation in stress responses, and on the finding that PIAL expression is stressresponsive, growth under a number of stressful conditions was tested for wild-type and mutant seedlings. Seeds were plated on phytagel medium, and the wet weight of germinated seedlings was determined. A total of 5468 plants were classified, more than 1200 for each condition (Supplemental Table 1). Figure 2A lists growth differences as seedling wet weight. Growth differences were particularly obvious under increased osmotic pressure (300 mM mannitol), upon addition of low doses of abscisic acid (ABA; 2.5 mM), and in presence of 150 mM NaCl. The pial1 pial2 double mutant grows less well under osmotic stress, and this trend is aggravated by the additional presence of a mutation in SIZ1. Likewise, pial1 and pial2 augment the ABA hypersensitivity phenotype of siz1 mutants. A surprising result was obtained in the salt stress experiment. Mutant pial2 and pial1 pial2 plantlets were significantly heavier than the wild type under these conditions ( Figure 2A). Similar differences were also observed regarding dry weight (Supplemental Table 1). Figure 2B and Supplemental Figure 8 show that pial1 pial2 double mutants are also greener in presence of 150 mM NaCl than wild-type plants. Chlorophyll fluorescence measurements were performed to relate this latter phenotype to photosystem II (PSII) performance (reviewed in Baker, 2008;Murchie and Lawson, 2013). From the fluorescence yield, two parameters were derived, the maximal quantum yield Fv/Fm commonly used to monitor the stress level in plants, and PSII operating efficiency Fq9/Fm9, which is an estimation of the fraction of absorbed light energy used for photosynthesis. Figure  3 shows that pial1, pial2, and pial1 pial2 mutants are significantly better able to maintain PSII activity under salt stress. Both Fv/Fm and Fq9/Fm9 of single mutants even approach the level of unstressed wild-type plants (;0.83 and ;0.6, respectively, under the light conditions of the experiment; Yin et al., 2012a), in accordance with their higher biomass and darker green color. Taken (A) Seedlings of the indicated genotypes were germinated and grown on MS plates containing either no additive, 300 mM mannitol, 2.5 mM abscisic acid, or 150 mM NaCl. Seedling fresh weight was determined for size comparison. Results for different mutant alleles did not differ and were pooled for the diagrams. Error bars show SE of the mean, and asterisks indicate significant difference from Columbia-0 reference grown at the same condition (P < 0.0001, two-sided t test). together, these results show that PIAL1 and PIAL2 affect stress responses. Interaction with SIZ1 may occur in response to ABA but is less likely for the osmotic or salt stress responses.

SUMO Conjugation in pial Mutants
We next were interested in global SUMO conjugate levels in mutant plants. Extracts from 2-week-old seedlings were subjected to gel blotting and probed with anti-SUMO1 antibody. Experiments shown in Supplemental Figure 9 indicated that, whereas the wild type and mutants differed in steady state levels of SUMO conjugates, conditions used to probe long-term effects on growth, in particular 300 mM mannitol treatment, gave results that were similar to unstressed samples. This contrasts with previous findings that SUMO conjugate levels change rapidly upon heat treatment (Kurepa et al., 2003). In order to also probe dynamic changes in SUMO conjugate levels, seedlings grown under normal conditions were compared with seedlings after a 1 h 37°C heat shock. Figure 4 shows a diagram of the results (four biological replicates). pial1 pial2 double mutants have similar or even higher SUMO conjugate levels than the wild type both under normal and under heat stress conditions. As previously published (Miura et al., 2005), siz1 mutants contain significantly less SUMO conjugates (one-sided t test, P < 0.05 and P < 0.005 for normal and heat stress conditions, respectively), and conjugate levels in siz1 mutants do not rise upon heat stress. In contrast, SUMO conjugate levels in the triple mutant pial1 pial2 siz1 under heat stress are close to wild-type levels (difference between siz1 and triple mutant at significance level P = 0.08). To interpret this surprising result, we hypothesize that SUMO conjugation by PIAL1 and PIAL2 is actually part of a desumoylation route, in which PIAL activity to build SUMO chains on monosumoylated substrates leads to their removal, e.g., via SUMO chain-specific ubiquitin ligases (see Discussion). Lack of PIAL1 and 2 may therefore result in increased SUMO conjugate levels under some conditions.

Assessment of Metabolite Differences
Changes in metabolites were also associated with mutations in SUMO ligases. We compared siz1 mutants to pial1 pial2 mutants (two different alleles) and to wild-type plants grown on soil under standard greenhouse conditions. Exact values for all metabolites are listed in Supplemental Table 2. Figure 5 shows that siz1 and pial1 pial2 mutants differ from the wild type and from each other in the content of many metabolites. For instance, whereas glucose levels in pial1 pial2 mutants are comparable to the wild type, glucose content is decreased in siz1. Conversely, fructose is elevated in pial1 pial2, but close to the wild type in siz1. Malate content in pial1 pial2 is in the wild-type range but elevated in siz1. Statistical analysis clearly puts pial1-1 pial2-2 and pial1-3 pial2-1 double mutants close to each other, and distinct from siz1, confirming equivalence of T-DNA insertion alleles and the hypothesis that PIALs and SIZ1 perform distinct functions. Measurement of nitrate content further supports this notion ( Figure 6A). Nitrate reductase (NR) is a known substrate of SIZ1-dependent SUMO conjugation and is activated by sumoylation (Park et al., 2011). Increased levels of nitrate in siz1 mutants are consistent with Plants were grown on plates containing 150 mM NaCl as in Figure 2. Fluorescence images of 2-week-old seedlings were recorded, and ratios Fv/Fm, as well as Fq9/Fm9, were calculated as described in Methods. Results indicate increased PSII performance in pial1 and pial2 as well as in pial1 pial2 double mutants. Asterisks indicate significant differences to the wild type (**P < 0.0001; *P < 0.05; two-sided t test). Protein extracts from 2-week-old seedlings were subjected to protein blot and hybridization to anti-SUMO1 antibodies and to protein determination for normalization (Supplemental Figure 9). Results were quantitated to obtain bar diagrams, using Columbia-0 without stress treatment as standard. Plants were subjected to 1 h 37°C heat shock or left at 23°C prior to protein harvest. Error bars indicate SE of the mean of four biological replicates. Significance level: *P < 0.05 and **P < 0.005 (one-sided t test, comparison of siz1 mutant to the wild type of similar treatment). reduced NR activity, as reduced activity may lead to substrate accumulation. In contrast, pial1 pial2 mutants have nitrate levels slightly below the wild type, suggesting that the influence of PIALs on NR activity may slightly oppose that of SIZ1. The analysis also showed deviations from the wild type in sulfate content, which is particularly pronounced for pial1 pial2 mutants. We therefore included transcription of key genes of sulfur transport and metabolism in the analysis, using quantitative RT-PCR with primers and gene identifiers listed in Supplemental Table 3. We find that most of the genes of this category are downregulated in pial1 pial2 mutants ( Figure 6B). Figure 5 shows that Met, one of the key sulfurcontaining metabolites, is lowered in SUMO conjugation mutants. Figure 7 displays data for another set of sulfur-containing key metabolites, Cys and the Met precursor homocysteine (Homocys), as well as GSH and its precursor g-glutamyl-cysteine (g-EC). Whereas Met is decreased, Cys is significantly elevated in pial1 pial2 mutants. Because Homocys is not elevated, these mutants may have less capacity to convert Cys to Homocys, and subsequently to Met. The tentatively elevated Thr content and the significantly increased contents of Ser, the precursor of Cys, support this interpretation. GSH content is elevated in pial1 pial2 mutants, as a result of the increased Cys content. GSH precursor g-EC is also elevated, completing the picture of increased sulfur flow into this redox protective branch. In contrast, siz1 mutants contain more Homocys, and their decreased Met content may be related to a less efficient conversion of Homocys to Met or higher turnover via S-methylmethionine and S-adenosylmethionine. GSH levels are closer to normal than in the pial1 pial2 mutants. In both mutants, however, regulation of sulfur assimilation is apparently perturbed. The concentration of key metabolites was determined to assess differences in SUMO conjugation mutants. Color coding was performed according to the indicated log 2 -scale heat map. (A) Concentration of nitrate and sulfate in leaves. Asterisks indicate significant difference from the wild type (P < 0.05). (B) Expression level of key genes related to nitrate and sulfate metabolism, as determined by quantitative RT-PCR, color-coded according to the log 2scale shown below. ATPS, ATP sulfurylase; AtPt, phosphate transporter; Sultr, sulfate transporter; APSK, APS kinase; APSKput, putative APSK; APR, APS reductase; NRT, nitrate transporter; Nitrite red, nitrite reductase.

In Vitro Activity of PIAL Proteins
In order to learn more about the biochemical reactions catalyzed by PIAL1 and PIAL2, full-length and shortened proteins were expressed in Escherichia coli cells and used for in vitro SUMO conjugation reactions, together with E. coli-produced Arabidopsis SAE and SUMO conjugating enzyme (SCE1) (Budhiraja et al., 2009). Figure 8 shows some of the constructs analyzed. We noticed that both PIAL1 and PIAL2 enhanced the normally moderate ability of SCE1 to produce SUMO chains. The activity is displayed by full-length PIAL1 ( Figure 8) and PIAL2 (Supplemental Figure 10) and by fragments containing the SP-RING domain together with surrounding protein sequences (amino acids 264 to 445 for PIAL1, called PIAL1M in the following, and 281 to 496 for PIAL2, called PIAL2M). The SP-RING domain alone was catalytically inactive for both proteins. SUMO chain formation is enhanced by PIAL proteins both at low and at high concentration of SCE1 (shown for PIAL2M in Figure 9A; the darkness of the SUMO chain area of the blot was quantified as described in Methods). Furthermore, PIAL2M also enhanced chain formation by SUMO3 ( Figure 9B). Mass spectrometry analysis of SUMO1 chains indicated that the same Lys residues were used for SUMO-SUMO bonds as found in reactions with SCE1 alone, namely, Lys-10, Lys-23, and Lys-43. Use of a broad spectrum of SUMO isoforms by PIAL proteins could also be demonstrated by detection of in vitro autosumoylation, which occurred with SUMO1, SUMO3, SUMO5, and SUMO7 (Supplemental Figures 11 and 12). We want to emphasize that autosumoylation of a SUMO ligase in vitro, which generally occurs with any active ligase, may not have in vivo significance. In the context of this work, this reaction supports the notion that PIAL ligases can productively interact with SCE1, which carries a broad variety of SUMO isoforms.
To define functional domains, mutants of the middle sized PIAL2 fragment PIAL2M were tested. Figure 10 and Supplemental Figure 13 show that this protein contains several redundantly acting sequences. The SP-RING motif was mutated by changing two of the Zn 2+ coordinating Cys residues (shown in bold blue in Supplemental Figure 11A) to Ala. The experiment showed that, surprisingly, destruction of SP-RING functionality does not completely abolish chain-forming activity. Furthermore, a bioinformatic search for potential SUMO interaction motifs (SIMs; Hecker et al., 2006;Vogt and Hofmann, 2012) suggested sequences VFDL and IFDI as candidates (shown as yellow bars in Figure 10A and as red letters in Supplemental Figure 11A). SIMs are short exposed hydrophobic/acidic sequences that can bind to SUMO. The two sequences were mutated either singly, or both to Ala stretches (AAAA; red bars in Figure 10A). Mutation of SIM1 led to a slightly changed spectrum of SUMO chains, and activity of the mutant protein was higher under some experimental conditions (Supplemental Figure 14). However, combined inactivation The concentration of Cys, Homocys, GSH, and g-EC was determined in wild-type and mutant leaves. Asterisks indicate significant difference from the wild type (P < 0.05).
of the SP-RING and motif VFDL (SIM1) resulted in complete loss of activity, suggesting that SIM1 and the SP-RING act in part redundantly to catalyze chain formation. A test of deletion variants of PIAL2M (Supplemental Figure 13) showed that deleting the sequence N-terminal to the SP-RING reduced activity and combining this deletion with a SIM1 mutation also led to complete loss of in vitro activity. For mutant protein comparison, two hour in vitro reactions were used as a standard, as shorter time points offer less sensitivity, whereas longer time points may exceed the range of linearity (Supplemental Figure 14).
Analysis of in vitro sumoylation products on Coomassie bluestained gels indicated two bands that were more prominent in reactions containing PIAL2M than in reactions with SAE and SCE1 only. Mass spectrometry analysis indicated that these bands represent SCE1 linked to SUMO via Lys-15 of SCE1. Immunoblotting with anti SCE1 antibody confirmed this analysis ( Figure 11A; the lower SCE1-SUMO band resulted from proteolytic loss of the N-terminal extension that was appended to SUMO for in vitro detection purposes). In order to find out whether SUMO conjugation to Lys-15 of SCE1 is functionally important   5 and 9, 1.75 mM). Reactions of lanes 6 to 9 contained 1.5 mM PIAL2M. Reaction products were gelseparated and transferred to a membrane, and SUMO was detected immunologically. Bottom, result of SUMO chain abundance measurements from the blot above (line to the right of the blot indicates size range used for quantification). for SUMO chain formation, an SCE1 K15R mutant was prepared. It could not form significant amounts of SUMO chains in the presence or absence of PIAL2M ( Figure 11B). In contrast, its activity to sumoylate the model substrate Nucleosome Assembly Factor (Budhiraja et al., 2009) in vitro was unchanged ( Figure  11C). We therefore conclude that SUMO conjugation to SCE1 at position Lys-15 is essential for SUMO chain formation in Arabidopsis. Interestingly, a highly active mutant version of PIAL2M (SIM1 Ala substitution mutant of Figure 10 and Supplemental Figure 14) was still able to generate low amounts of SUMO chains with SCE1 K15R. This result may indicate that PIAL ligases have a direct role in SUMO chain formation (preferentially together with SUMO-modified SCE1, but also, less efficiently, with nonmodified SCE1) and that the role of PIALs in chain formation is not restricted to enhanced generation of SUMO-modified SCE1. Nonetheless, SUMO-modified SCE1 is capable of forming SUMO chains without PIALs (its low level presence is apparently responsible for the moderate amounts of SUMO chains formed in absence of PIALs), although it works less efficiently than SUMO-SCE1 and PIAL together.

DISCUSSION
In this work, we characterized two Arabidopsis proteins: PIAL1 and PIAL2. They contain an SP-RING domain, which is frequently present in SUMO ligases, and are conserved in the plant kingdom (Novatchkova et al., 2012). Mutants in the respective genes are viable and grow normally under standard greenhouse conditions but show differences under environmental stress conditions. Most remarkably, pial1 pial2 mutants grow better than the wild type in the presence of 150 mM NaCl (two different T-DNA insertion alleles of both pial1 and pial2 were used to exclude effects of potentially present unintended mutations). They are darker green, their PSII performance is close to that of nonstressed wild-type plants, and they have higher fresh weight when grown on plates (Figures 2 and 3). In contrast, these plants are less able to maintain growth under osmotic stress (300 mM mannitol; Figure 2). It remains to be seen how these properties relate to soil-grown plants, which may encounter both stresses at the same time. The SUMO conjugation system has previously been linked to salt stress response, via SUMO-specific proteases OVERLY TOLERANT TO SALT1 (OTS1) and 2 (Conti et al., 2008). Overexpression of either of these two proteases renders Arabidopsis plants more resistant to salt stress. OTS and PIAL may therefore act in the same pathway of salt tolerance.
In order to compare PIAL1 and 2 to the known SUMO ligase SIZ1, we generated double and triple mutants. Their analysis established that loss of PIALs in the siz1 background aggravates the already existing hypersensitivity of siz1 to the stress hormone ABA (Figure 2A). Double and triple mutant combinations also reduce growth of siz1 mutants on soil (Supplemental Figure 7), suggesting that siz1 mutants are particularly sensitive to loss of PIAL function. The analysis of SUMO conjugate levels in the mutants (Figure 4) indicated that mutants in PIAL 1 and 2, in contrast to mutants in SUMO ligase SIZ1, have normal or slightly elevated SUMO conjugate levels. However, combining siz1 with pial1 and pial2 mutations largely reverses the reduced SUMO conjugate levels of siz1 under heat shock. This is an unusual result for removal of SUMO ligases, for which one might expect a further reduction in SUMO conjugate levels. Although part of this effect might result from compensatory changes in activity of the remaining sumoylation components, we favor an explanation for this finding, as discussed below.
As a further characterization of mutants, we performed a detailed metabolite analysis. Figure 5 indicates a profound difference between pial1 pial2 and siz1 mutants. For instance, siz1 had reduced glucose content, whereas pial1 pial2 had normal glucose levels. Malate was increased in siz1 and normal in pial1 pial2, whereas fructose was normal in siz1, but increased in pial1 pial2. A detailed list of all metabolites is presented in Supplemental Table 2. Figure 6A examines two major nutrient ions: nitrate and sulfate. siz1 mutants had increased nitrate content, whereas the content was slightly lowered in pial1 pial2 plants. The increase in siz1 can be related to the finding (Park et al., 2011) that sumoylation of (A) Scheme of constructs and relative activity. In addition to the SP-RING, two potential SIMs were mutated in PIAL2M. Yellow bars indicate intact functional elements and red bars mutated elements. The abundance of SUMO chains in the blot of (B) is shown to the right of each construct (values relative to the reaction without PIAL2M protein). (B) Activity assay (anti SUMO protein gel blot) used to determine relative activity values of (A). Bottom insert shows section of a Coomassie bluestained gel to document the amount of PIAL2M variant protein added to the reaction. nitrate reductase by SIZ1 activates this enzyme. It is thus plausible that the lack of efficient nitrate reduction is responsible for the accumulation of nitrate. This interpretation is supported by the finding that siz1 and siz1 pial double and triple mutant combinations grow better in presence of additional ammonium ions (Park et al., 2011; data not shown). We conclude from the data that PIAL1 and PIAL2 have only minor influence on nitrate metabolism. The second diagram of Figure 6A shows sulfate levels. The levels were significantly elevated in pial1 pial2 double mutants, but much less so in siz1 mutants.
An influence of SUMO conjugation on sulfate metabolism was not previously reported. We therefore gathered more data on components of sulfur-related pathways. Although SUMO conjugation is a posttranslational modification, its perturbation may well affect gene expression. Figure 6B shows gene expression data on sulfate transporters and on key enzymes of sulfur metabolism. Even a first glance leads to the conclusion that there is significant deviation from wild-type expression. Consistent with the more significant deviation of sulfate levels in pial double mutants, less gene expression differences are found in the siz1 mutant. Most genes of sulfur metabolism are downregulated in pial1 pial2 mutants. Figure 5 demonstrates decreased levels of Met in both pial1 pial2 double mutants and in siz1 single mutants. We determined concentrations of further key sulfur containing compounds and their precursors. In particular, Figure 7 shows that pial1 pial2 mutants had elevated Cys levels. In contrast, an intermediate of the Cys to Met conversion, Homocys, is not elevated in pial1 pial2 mutants. Future investigations may therefore capitalize on the step from Cys to Homocys as potentially requiring PIAL assistance. siz1 mutants, in contrast, have significantly increased Homocys levels, so that steps from Homocys to Met may require SIZ1, or enzymes of the Yang cycle might be regulated by sumoylation (Sauter et al., 2013). GSH, as a major component of redox balance, was also investigated in detail. pial1 pial2 plants had elevated levels, whereas siz1 mutants did not significantly deviate from the wild type. Thus, the increased flow of Cys into GSH might be an effect of the deregulated Met formation, usage via the Yang cycle, or the deregulation of the enzyme methionine-g lyase, degrading Met (Jander and Joshi, 2010).
In vitro SUMO conjugation was chosen as a method to characterize PIAL1 and PIAL2 biochemically. All components of the in vitro system were expressed and purified from E. coli, based on Arabidopsis cDNA sequences (Budhiraja et al., 2009). Figure 8 and Supplemental Figure 10 show that, most significantly, both  (lanes 2 and 4). Dots, SUM1-SCE1 (higher band with intact extension on SUMO1; lower band with proteolytically shortened extension); asterisk, cross-reacting band.
(B) SCE1 K15R has a drastically reduced ability to form SUMO chains (lanes 7, 9, and 11 versus 6, 8, and 10). Lane 5 shows SUMO without SAE and SCE1. (C) SCE1 K15R can catalyze SUMO conjugation to model substrate nucleosome assembly factor (NAF) as efficiently as SCE1 (lanes 14 and 16 versus 15 and 17), and (mono)sumoylation of this substrate is not enhanced by the presence of PIAL2M. Image shows anti NAF blot, with unmodified NAF as the major band. Dot at the right margin indicates NAF-SUMO band. (D) Peptide backbone and selected side chains of Arabidopsis SCE1, modeled from structural data of human UBC9. C94 is the active site Cys residue (yellow), NH 2 indicates the N-terminal end of the peptide chain, and Lys-15 (red) is the Lys residue used for SUMO addition. a-Helical parts are in pink, and b-sheets are in light blue. The rest of the protein backbone is in gray. PIAL1 and PIAL2 increase the extent of SUMO chain formation in vitro. This property was also characteristic of smaller protein fragments, consisting of the SP-RING domain plus surrounding sequences. The SP-RING alone, in contrast, had no significant activity. The construct containing SP-RING plus surrounding sequences of PIAL2 was used for further characterization (Figures 9  and 10; Supplemental Figures 13 and 14). SUMO chain enhancement was evident both at low concentrations of SUMO conjugating enzyme SCE1 and at high concentrations.
In vitro mutagenesis showed that an intact SP-RING domain is not absolutely essential for SUMO chain-forming activity. Chainforming activity is decreased, but not abolished, in a mutant with two changes in functionally critical, Zn 2+ -coordinating residues ( Figure 10). A bioinformatics search for additional functional domains identified two potential SUMO interaction motifs (SIMs; called SIM1 and SIM2 in this work), which consist of exposed hydrophobic and acidic residues that mediate SUMO binding. Most likely due to potential redundancy of SUMO binding sites in PIAL2M, which resulted in a high background, we were not able to directly demonstrate that the identified motifs bind to SUMO. However, simultaneous mutation of both SP-RING and SIM1 completely eliminated activity of PIAL2M. Likewise, deletion of the protein part N-terminal to the SP-RING decreased activity and in combination with the SIM1 mutation led to complete elimination of activity (Supplemental Figure 13). We interpret that PIAL2M contains an array of redundantly acting functional elements, and the SP-RING is just one of them.
Testing of PIAL-mediated enhancement of SUMO conjugation to a number of previously identified substrates (Budhiraja et al., 2009) was unsuccessful, suggesting that PIALs may be specialized in extending single SUMO residues on substrates into chains. However, we noticed that PIAL2M catalyzes increased SUMO conjugation to SCE1 ( Figure 11A). Mass spectrometry analysis indicated conjugation onto Lys-15 of SCE1. The SCE1 K15R mutant was used to analyze the role of this conjugate in SUMO chain formation. Figure 11B shows that SCE1-(K15)-SUMO is actually essential for chain formation. This poses the perplexing possibility that the role of PIAL in SUMO chain formation could actually be restricted to formation of SUMO-(K15)-SCE1, which then in turn forms SUMO chains all by itself. However, Figure 11B (lane 11) argues against this hypothesis and for a direct role of PIALs, in that high concentrations of PIAL2M (sim1 mutant) were capable of enhancing SUMO chain formation even with the SCE1 K15R mutant. Consistent with this model is the finding that PIAL2M catalyzes chain formation with SUMO3, which is a poor chain former in absence of PIALs ( Figure 9B). In any case, an important step in PIAL activity is to enhance SUMO ligation to SCE1. SUMO-SCE1 could act instead of SCE1 as a SUMO donor (by carrying a second SUMO, for transfer to substrates, on its active site Cys) or act as a cofactor in addition to unmodified SCE1 as the SUMO carrier. That Lys-15 of SCE1 might be dispensable for normal SUMO conjugation activity was demonstrated by the reaction shown in Figure 11C, which indicates that monosumoylation of a model substrate occurs with equal efficiency by both SCE1 and the SCE1 K15R mutant. Figure 11D shows the critical Lys-15 residue in an SCE1 structure model, indicating that SUMO conjugation to this residue does not obstruct the active site cleft. SUMO conjugation to SCE was previously shown in mammals and in yeast, and it changes catalytic properties and "ligase-like" functions of SCE1 (Knipscheer et al., 2008;Klug et al., 2013). The SUMO-modified yeast enzyme participates in SUMO chain formation (Klug et al., 2013). It appears that in plants, additional dedicated SUMO ligases PIAL1 and PIAL2 control and enhance such a chain-forming activity. Because exclusively chain-forming ligases were called E4 enzymes in the ubiquitin conjugation system, we propose to adopt this terminology for PIAL proteins as well.
Which roles can be imagined in vivo for SUMO chains? We propose that SUMO chain formation can lead to removal of SUMO substrates and substrate assemblages. Although the data presented do not allow conclusions about the quantitative importance of such a pathway, this model can explain why pial1 pial2 mutants contain more SUMO conjugates under certain conditions (e.g., in heat-shocked siz1 genetic background). In particular, SUMO chains on substrates generate binding sites for a class of ubiquitin ligases and thus channel them into the ubiquitin-proteasome degradation pathway. These latter enzymes have previously been described in animals and in yeast (Yin et al., 2012b;Sriramachandran and Dohmen, 2014). Similar proteins were predicted (Novatchkova et al., 2012) and experimentally demonstrated (Elrouby et al., 2013) in plants as well. Thus, as part of their in vivo activity of extending mono-SUMO into a chain, PIAL proteins may channel SUMO substrates into the ubiquitin pathway, connecting SUMO conjugation to the major proteolytic pathway of the cell.

Plant Material and Growth Conditions
Arabidopsis thaliana wild type and mutants were grown in a greenhouse as described (Hubberten et al., 2012); a 16-h photoperiod at a minimum of 200 mmol m -2 s -1 and a day/night temperature of 20°C/18°C were used. For metabolome analysis, plant material was harvested 5 weeks after sowing in soil. Plant material was pooled, three plants per line and four pools per line. The experiment was repeated to give if possible three independent biological replicates. For seed propagation, plants were either grown in the greenhouse or in controlled environment chambers in 16-h-illumination/8-h-darkness conditions at 23°C. Mutant alleles used were: pial1-1, SALK_083748; pial1-3, SAIL_738_B09; pial2-1, SALK_043892; pial2-2, GK_712B09; and siz1, SALK_065397.

Growth on Plates
Arabidopsis seedlings were selected on plates containing Murashige and Skoog (MS) medium: 4.3 g/L MS salts, 0.5 g/L MES, 10 g/L sucrose, and 3.2 g/L GelRite, pH 5.7; MS vitamin mix (Sigma-Aldrich) was added after autoclaving. For continuous stress, the plates were supplemented with 300 mM mannitol, 150 mM NaCl, or 2.5 mM ABA, and the seedlings were grown for 2 weeks in long-day conditions. For heat shock stress, plants were grown on MS plates without stress factors for 2 weeks in long-day conditions. The seedlings were transferred to 1 mL liquid MS medium in a 24-well plate, one plant per well. After 24 h, the 24 well plates were floated in a 37°C water bath. After the stress, seedlings were frozen in liquid nitrogen and ground using a Qiagen Tissue Lyser II for 33 1 min at 30 shakes/second. Then, 200 mL Buffer B+ (90 mM HEPES, 30 mM DTT, 2% [w/v] SDS, 20 mg/mL pepstatin, and one tablet protease inhibitor cocktail [Roche] per 7 mL buffer, pH 7.5) were added, and the samples were heated for 6 min at 95°C. Then they were centrifuged for 5 min at 15,000g, and the supernatant was used in immunoblot analysis.

RNA Detection
For RT-PCR results shown Figure 1 and Supplemental Figure 6, mRNA was isolated using the RNeasy Plant Mini Kit (Qiagen) with on-column DNase digestion according to the manufacturer's protocol. First-strand DNA synthesis was performed with 140 ng total RNA using the Life Technologies Super Script II reverse transcriptase kit according to the manufacturer's protocol. One microliter of the first-strand reaction was used as a template in a standard PCR with Promega G2 polymerase for 30 cycles. Every reaction was performed in three technical replicates; UBC9 (At4g27960) was used as a housekeeping gene for comparison. PCR reactions were run for 33 cycles. Primers are listed in Supplemental Table 3.

Chlorophyll Fluorescence Measurements
Plants were grown on plates containing 150 mM NaCl for 3 weeks under long-day conditions at 120 mmol photons m 22 s 21 . Chlorophyll fluorescence was detected by an Imaging-PAM fluorometer (Walz). Before measurement, plants were incubated in darkness for 10 min. After turning on the measuring light, a saturating pulse (SP; 8000 mmol photons m 22 s 21 ) was applied to determine the Fv/Fm value. Subsequently, an actinic light (134 mmol photons m 22 s 21 ) was turned on and a sequence of SPs was applied (one SP every 30 s) to monitor Fq9/Fm9 (PSII operating efficiency) under steady state illumination.

Vector Construction
A list of vectors is provided as Supplemental Table 4. Standard cloning techniques were used for vector construction. Site-directed mutagenesis was performed with the Agilent QuikChange II kit. For PIAL1 cDNA isolation, seedlings were germinated in liquid half-strength MS medium (six-well microtiter plates). H 2 O 2 was added to 10 mM 1 h prior to plant harvest. RNA was extracted (Qiagen RNeasy Plant Mini Kit) and used for reverse transcription (Invitrogen Superscript III) and PCR amplification (TaKaRa LA Taq).

Protein Expression and Purification
Recombinant Arabidopsis proteins were expressed in Escherichia coli using the Rosetta strain (BL21 DE3 pLysS), induced with 1 mM isopropyl b-D-1thiogalactopyranoside. SUMO isoforms had an N-terminal extension with a poly-His tag (Budhiraja et al., 2009) and were purified using a Ni 2+ -based affinity chromatography (Qiagen) according to the manufacturer's manual, with the binding buffer supplemented with 10% (v/v) glycerol and 0.5% (v/v) Triton X-100. PIAL1 and PIAL2 variants were N-terminally tagged with maltose binding protein and were purified with an amylose resin (NEB). The two subunits of the E1 (SAE1 and SAE2) were expressed in stoichiometric amounts from a dicistronic construct (Budhiraja et al., 2009). Only SAE2 carried a Histag. During purification with Ni resin, all buffers were supplemented with 5 mM ATP to promote complex formation. SCE1 was expressed without tag and was extracted by ultracentrifugation of bacterial pellets at 100,000g for 1 h after subjecting the pellets to freeze-thaw cycles (Tomanov et al., 2013).

Immunoblotting and Protein Detection
Proteins were separated using SDS-PAGE and transferred to a polyvinylidene fluoride membrane (Immobilon-P, pore size 0.45 mm; Millipore) by wet blotting for 1 h (1.5 h in case of very large proteins) at 50 V and 4°C. Afterwards, the membrane was blocked for 2 h in 5% skimmed milk powder in Tris-buffered saline (TBS) + 0.05% Tween 20 at room temperature and then washed once with TBS. The primary antibody was diluted in TBS and applied to the membrane overnight at 4°C. The membrane was washed twice with TBS + 0.05% Tween 20 for 5 min and once with TBS only. For conjugated antibodies, the membrane was developed immediately. For nonconjugated antibodies, the secondary antibody was diluted in TBS and applied on the membrane for 2 h at 4°C. Antibodies used were: rabbit anti-Arabidopsis SUMO1 (Agrisera; used in 1:2000 dilution), monoclonal mouse anti-maltose binding protein (New England Biolabs; 1:10,000 dilution), AP-conjugated streptavidin (IBA; 1:10,000 dilution), rabbit anti-SCE1 (Budhiraja et al., 2009; 1:100 dilution). After the secondary antibody incubation, the membrane was again washed twice with TBS + 0.05% Tween 20 for 5 min and once with TBS only. For alkaline phosphatase-conjugated secondary antibodies (Sigma-Aldrich anti mouse-IgG or anti rabbit-IgG, both in 1:30,000 dilution): 66 mL 5% Nitro Blue Tetrazolium (Sigma-Aldrich) in 70% N,N-dimethylformamide and 33 mL 5% 5-bromo-4-chloro-3-indolyl phosphate (Gerbu) in 100% N,N-dimethylformamide were added to 10 mL TE buffer and applied on the membrane in the dark. For horseradish peroxidase-conjugated antibodies (horseradish peroxidase-conjugated anti rabbit IgG, as provided in the Amersham ECL protein gel blotting kit; GE Healthcare), the membrane was incubated for 2 min in the dark with either Amersham ECL protein gel blotting detection reagents (GE Healthcare) or WesternBright Sirius detection reagents (Pierce). A Fuji medical x-ray film (Fujifilm) was exposed to the membrane and subsequently developed.

Quantitation of Immunoblot Results
For Figure 4 and Supplemental Figure 9, every sample was separated on two gels, one of which was stained with Coomassie Brilliant Blue R 250, and the other was analyzed by immunoblotting. The gels and membranes were scanned and the intensity of their staining was quantified using ImageJ. The anti-SUMO immunoblot intensities were normalized to the Coomassie stain intensity. Samples were compared with the nonstressed Columbia-0 control, which was assigned a value of 1. For the in vitro sumoylation tests (Figures 9 and 10), membranes were scanned and quantified using ImageJ. Sample intensities were compared with the reaction without ligase fragment, which was assigned a value of 1.

Image Processing (SCE1 Structure Data)
The three-dimensional structure of SCE1 was modeled with the ESy-Pred3D prediction engine (Lambert et al., 2002) using the human homolog UBC9 (PDB ID 2GRN). The three-dimensional structure of the PIAL2 SP-RING (Supplemental Figure 3) was modeled using the bakers' yeast SIZ1 (PDB ID 3E2D).

Metabolite Profiling Using Gas Chromatography Coupled to Mass Spectrometry
For gas chromatography-mass spectrometry analysis, polar metabolite fractions were prepared from 50 mg frozen leaf tissue, which was ground to a fine powder and then extracted with methanol/CHCl 3 . The metabolite extraction and the preparation of the fraction of polar metabolites by liquid partitioning into water were described earlier (Roessner et al., 2000;Wagner et al., 2003). Metabolite samples were derivatized by methoxyamination, using a 20 mg/mL solution of methoxyamine hydrochloride in pyridine, followed by a subsequent trimethylsilylation step, with N-methyl-N-(trimethylsilyl)-trifluoroacetamide (Fiehn et al., 2000;Roessner et al., 2000). A C12, C15, C19, C22, C28, C32, and C36 n-alkane mixture was used for the determination of retention time indices, and chromatographic alignment (Wagner et al., 2003). Ribitol, isoascorbate, and deuterated alanine were added for internal standardization (Kaplan et al., 2004). Samples were injected using splitless mode (1 mL/sample) and analyzed using a quadrupole-type gas chromatography-mass spectrometry system (MD800; ThermoQuest). The chromatograms and mass spectra were evaluated using the MASSLAB software (ThermoQuest), and data were processed using the software TagFinder (Luedemann et al., 2008). Following the procedure of Desbrosses et al. (2005) peak areas, X, were defined to represent the fragment responses (X i , of fragment i). Fragment responses were normalized to fresh weight and response of the internal standard ribitol using fragment m/z = 319, at RI = 1733.7.

Extraction and Metabolite Analysis
Individual soluble thiols were determined as the sum of their reduced and oxidized forms. Fifty milligrams of frozen ground leaf tissue were added to 25.0 mg polyvinylpolypyrrolidone (previously washed with 0.1 M HCl) and 500 mL of 0.1 M HCl. The samples were shaken for 60 min at room temperature. After centrifugation (15 min at 15,777g, 4°C), the supernatants were frozen at 220.0°C until reduction/derivatization. The levels of glutathione and cysteine were determined by HPLC-based method after subsequent reduction and derivatization with monobromobimane, as described by Kreft et al. (2003).

Quantitative Real-Time PCR
The extraction of total RNA was performed with use of Trizol reagent (Invitrogen). RNA from plant leaves was extracted in 700 mL of the reagent, and after addition of 200 mL chloroform, the polar phase was transferred into a new tube. The aqueous phase was mixed with isopropanol (0.6 volumes) and 3 M sodium acetate (0.2 volumes) to precipitate RNA at room temperature. The RNA precipitate was washed three times with 70% ethanol. The aqueous solution of RNA was treated with DNase (Thermo Scientific) according to the manufacturer's protocol. The cDNA was synthesized from DNA-free RNA by use of oligo-dT 18 and the RevertAid Premium Reverse Transcriptase (both Thermo Scientific) according to the manufacturer's protocol. PCR was conducted in an optical 384-well plate with an ABI PRISM 7900 HT sequence detection system (Applied Biosystems). Reactions contained 5 mL 2-times SYBR Green Master Mix reagent (Applied Biosystems), 1 mL of 1:5 diluted shoot, 1 mL 1:2.5 diluted root cDNA, and 200 nM of each gene-specific primer. Total reaction volume was 10 mL. The thermal profile used was 50°C for 2 min, 95°C for 10 min, 40 cycles of 95°C for 15 s, and 60°C for 1 min. SDS 2.0 software (Applied Biosystems) was used for data analysis. CT values for genes were normalized to the CT values of ubiquitin. PCR efficiencies were calculated using the LinRegPCR 7.0 program (Ramakers et al., 2003). All primer sequences were designed using the criteria described by Czechowski et al. (2004). RT primer sequences are listed in Supplemental Table 3.

Statistical Analysis
Heat map presentation was performed on the data sets obtained from metabolite profiling with the software package TMEV (Saeed et al., 2003).
The data were log 2 transformed before analysis. The t tests and error bars in figures were calculated using the algorithm embedded in Microsoft Excel together with add-on Daniel's XL Toolbox.

Supplemental Data
The following materials are available in the online version of this article.