Proteomic analysis of affinity-purified 26S proteasomes identifies a suite of assembly chaperones in Arabidopsis

The 26S proteasome is an essential protease that selectively eliminates dysfunctional and short-lived regulatory proteins in eukaryotes. To define the composition of this proteolytic machine in plants, we tagged either the core protease (CP) or the regulatory particle (RP) sub-complexes in Arabidopsis to enable rapid affinity purification followed by mass spectrometric analysis. Studies on proteasomes enriched from whole seedlings, with or without ATP needed to maintain the holo-proteasome complex, identified all known proteasome subunits but failed to detect isoform preferences, suggesting that Arabidopsis does not construct distinct proteasome sub-types. We also detected a suite of proteasome-interacting proteins, including likely orthologs of the yeast and mammalian chaperones Pba1, Pba2, Pba3, and Pba4 that assist in CP assembly; Ump1 that helps connect CP half-barrels; Nas2, Nas6, and Hsm3 that assist in RP assembly; and Ecm29 that promotes CP–RP association. Proteasomes from seedlings exposed to the proteasome inhibitor MG132 accumulated assembly intermediates, reflecting partially built proteasome sub-complexes associated with assembly chaperones, and the CP capped with the PA200/Blm10 regulator. Genetic analyses of Arabidopsis UMP1 revealed that, unlike in yeast, this chaperone is essential, with mutants lacking the major UMP1a and UMP1b isoforms displaying a strong gametophytic defect. Single ump1 mutants were hypersensitive to conditions that induce proteotoxic, salt and osmotic stress, and also accumulated several proteasome assembly intermediates, consistent with its importance for CP construction. Insights into the chaperones reported here should enable study of the assembly events that generate the 26S holo-proteasome in Arabidopsis from the collection of 64 or more subunits.

formed, at which point the two halves assemble and Ump1 is degraded, thus becoming the first substrate of the complex (22).
Correct assembly of the RPT ring of the RP also occurs with the help of five dedicated chaperones, called Nas2, Nas6, Hsm3, Rpn14, and Adc17 in yeast (or p27, p28, S5b, and PAAF1, respectively, in mammals, which lack an Adc17 homolog (5,13,14)). These chaperones independently bind to the C-terminal region of specific RPT subunits, resulting in three distinct precursor assembly modules (13,14). These modules associate sequentially, followed by incorporation of RPN2, RPN13, and RPN10 to finalize RP base assembly and trigger chaperone eviction (13). As of yet, no assembly chaperones have been identified for the RP lid, which is currently thought to combine spontaneously using the RPT ring as a scaffold, based on selfassembly studies in vitro and in Escherichia coli (13,25).
The final step in 26S proteasome assembly is association of the RP with the CP, which is driven by docking of the C-terminal HbYX motifs from several RPT ring subunits onto the ␣-subunit rings of the CP (11). This association also occurs spontaneously in vitro, is stabilized by ATP, and can be fully reversible (4,5). Here, Ecm29 appears to provide a critical quality control checkpoint by binding to structurally aberrant proteasomes and repressing both the ATPase activity of the RP and gate opening of the CP (26,27).
Our understanding of the 26S proteasome composition in multiple species has been aided by the use of tagged subunits for its rapid affinity purification, which, when followed by deep proteomic analysis, allowed identification of all known CP and RP subunits and a large collection of proteasome-associated proteins (28,29). Using this strategy, we attempted to better define the composition of the Arabidopsis thaliana complex, with a focus on proteins that interact with the 26S particle and thus might be important for assembly and/or regulation. As most subunits of the Arabidopsis complex are encoded by gene pairs, we also tested the possibility that plants, like mammals, assemble multiple proteasome isoforms with potentially unique activities, which are created by the deliberate pairing of subunit paralogs. Although our mass spectrometry (MS) results failed to find evidence for unique 26S proteasome isotypes, we did discover numerous factors that bind to the CP and/or RP sub-complexes. Included in this list were orthologs for a number of yeast and mammalian proteasome chaperones needed for CP and RP assembly, the CP effector PTRE1 (known as Fub1 and PI31 in yeast and humans, respectively (30)), and the PA200/Blm10 regulator that caps the CP (29). Genetic analysis of the Arabidopsis UMP1 maturation factor revealed that it promotes CP assembly and, unlike its yeast counterpart (22), is essential. Taken together, this study represents the first description of plant proteasome assembly chaperones, and it provides a platform for future studies describing how this proteolytic machine is built in plants.

Generation of transgenic plants for RP-based proteasome purifications
Our prior proteomic studies on Arabidopsis 26S proteasomes relied on a transgenic line that genetically replaced the Identification of plant proteasome assembly chaperones single gene encoding the CP ␣-subunit PAG1 (␣ 7 ) with a FLAG-tagged variant (29). To complement this approach, we developed transgenic plants that replaced the paralogous RPT4a and RPT4b genes (At5g43010 and At1g45000, respectively (31)) encoding the RP base subunit RPT4 with 2ϫFLAGtagged versions that would allow us to specifically isolate the RP sub-complex or the full 26S holo-proteasome enriched for these isoforms. The two 399-amino acid RPT4 proteins (44.8 kDa each) share high sequence identity both to each other (97% identity between the two isoforms) and to orthologs from other plant, animal, and yeast species (Fig. S1, A-C). Importantly, the cryo-electron microscopic structure of the yeast proteasome showed that the N terminus of RPT4 is likely solventaccessible in the Arabidopsis particle (32), thus enabling addition of the 2ϫFLAG sequence without perturbing RP assembly and activity.
For recipients of tagged RPT4, we identified a collection of T-DNA insertion mutants within the RPT4a or RPT4b loci (Fig.  1A). Of use here were the rpt4a-1 and rpt4b-2 mutations, which, when homozygous, disrupted accumulation of the full-length transcripts as judged by RT-PCR analysis of total RNA, and thus they were considered to be null alleles (Fig. 1B). Surprisingly, homozygous rpt4a-1 and rpt4b-2 seedlings and mature plants, along with those harboring several other rpt4a or rpt4b alleles, were phenotypically indistinguishable from WT Col-0 plants when grown under nutrient-rich growth conditions and long-or short-day photoperiods (Fig. 1, C and D). This is in contrast to null mutants affecting several other RP subunit genes in Arabidopsis, which have been shown to display a range of phenotypic defects (33,34). However, the presence of at least one copy of RPT4 appeared to be essential to Arabidopsis, as we failed to identify double-homozygous rpt4a-1 rpt4b-2 progeny from self-crosses of double-heterozygous parents (Fig.  1E). Even homozygous/heterozygous progeny (Aabb or aaBb) could not be found, indicating that both the male and female haploid gametes require at least one copy of a functional RPT4 gene (Fig. 1E).
For the purification strategy used here, we complemented the rpt4a-1 and rpt4b-2 plants with FLAG-tagged versions of RPT4a and RPT4b, using a genomic fragment of each bearing  Fig. S1. Positions of oligonucleotide primers used for reverse transcription (RT)-PCR in B are indicated by the half-arrowheads. B, RT-PCR analysis of the RPT4a and RPT4b transcripts in the indicated T-DNA insertion mutants. Total RNA was extracted from 7-day-old WT or homozygous mutant seedlings, converted into cDNA, and subjected to RT-PCR using the indicated primer pairs as shown in A. Amplification of ACT2 was included to confirm analysis of near equal amounts of cDNA. C and D, rpt4a and rpt4b mutants described in A and B are phenotypically indistinguishable to WT. Shown are plants grown for 10 days in a long-day photoperiod (C), or for 42 days in a short-day photoperiod (D). E, RPT4 is essential in Arabidopsis. The genotypes of 230 progeny from a selfed double-heterozygous rpt4a-1/ϩ rpt4b-2/ϩ plant were determined by genomic PCR, and the number and percentage of each genotype observed are indicated, together with the expected percentages if all genotypes were viable. No double-homozygous plants were identified. F, Arabidopsis rpt4a-1 and rpt4b-2 complemented with the FLAG-RPT4a or FLAG-RPT4b transgenes are phenotypically indistinguishable to WT or their respective homozygous mutants. Shown are plants grown for 7 days on GM medium followed by 14 days on soil. G, anti-RPT4a antibodies equally recognize both RPT4a and RPT4b. Recombinant RPT4a-and RPT4b-bearing N-terminal His 6 tags were diluted in 5-fold increments, separated by SDS-PAGE, and immunoblotted with anti-RPT4a antibodies, using anti-His 6 antibodies to confirm near-equal protein loading. H, immunoblot analysis of total protein extracts from 7-day-old Arabidopsis seedlings with anti-RPT4 and anti-FLAG antibodies confirmed expression of the FLAG-RPT4a and FLAG-RPT4b proteins. Open and closed arrowheads locate tagged and untagged RPT4a/b, respectively. Immunodetection of histone H3 was used to confirm near equal-protein loading. Numbers on the right in panel (B) indicate size markers in basepairs, whereas those in panels (G) and (H) indicate molecular mass markers in kDa.
the codons for a 2ϫFLAG sequence appended after the initiator methionine codon and preceded by a 600-or 2,000-bp region harboring the native RPT4a or RPT4b promoters, respectively. Plants homozygous for FLAG-RPT4a rpt4a-1 or FLAG-RPT4b rpt4b-2 were then identified in the F 2 generation by genomic PCR and by the segregation ratio of BASTA tolerance in the F 3 generation, which was introduced into the plants as a selectable marker along with the transgenes. Notably, the complemented plants were phenotypically indistinguishable from both the homozygous rpt4a-1 and rpt4b-2 parents and WT Col-0, suggesting that the tagged subunits were not detrimental to Arabidopsis growth and development (Fig. 1F).
We then confirmed expression of the FLAG-tagged proteins by immunoblot analysis of crude extracts with anti-RPT4 antibodies, which were generated against the RPT4a isoform but equally recognize RPT4b, as judged by immunoblot analysis of recombinant proteins (Fig. 1G). Accumulation of slower-migrating FLAG-RPT4a or FLAG-RPT4b proteins was detected in the complemented plants with both anti-RPT4a and anti-FLAG antibodies, but not in non-transformed plants (Fig. 1H). In addition, bands representing untagged RPT4 were seen in samples from the transformed rpt4a-1 and rpt4b-2 plants, which were likely derived from the WT RPT4b and RPT4a loci remaining in each line, respectively (Fig. 1H). FLAG-RPT4a accumulation was consistently higher than that of FLAG-RPT4b across multiple independent transgenic lines, which was consistent with the normally higher levels of RPT4a versus RPT4b when comparing the levels of total RPT4 in the single rpt4a-1 and rpt4b-2 mutants (Fig. 1H).
For use as an affinity reagent, FLAG-RPT4a and FLAG-RPT4b should readily integrate into 26S proteasomes. Indeed, when we analyzed the assembly of 26S particles by glycerol gradient centrifugation of total protein extracts from Arabidopsis seedlings, we found that both tagged versions associated with the 26S particle along with other RP and CP subunits, with little to no free FLAG-RPT4a or FLAG-RPT4b species evident at the top of the gradients (Fig. 2). In the absence of phenotypic rescue, this integration also indirectly confirmed that the FLAG-tagged versions of RPT4 remain functional.
Notably, like mutants impacting other 26S proteasome subunits (33)(34)(35), the rpt4a-1 and rpt4b-2 plants without the FLAG-RPT4a or FLAG-RPT4b transgenes accumulated an array of sub-species representing free CP and RP, which were not evident in WT or the complemented lines (Fig. 2). The accumulation of these sub-species implied that the absence of RPT4a or RPT4b compromises efficient proteasome assembly in planta, possibly by an insufficient supply of total RPT4 either throughout the seedlings or in specific cells due to variations in expression patterns between the paralogs. Total protein extracts from 10-day-old wildtype (WT), rpt4a-1, FLAG-RPT4a rpt4a-1, rpt4b-2, and FLAG-RPT4b rpt4b-2 seedlings were subjected to glycerol gradient fractionation, and samples from each fraction were analyzed by SDS-PAGE followed by immunoblot with antibodies against the indicated proteasome subunits. The expected positions of free CP, free RP, and the holo-26S proteasome are indicated by the horizontal brackets. The location of each protein within the 26S complex is indicated for the WT gels. Numbers on the right indicate molecular mass markers in kDa. The FLAG-RPT4a and FLAG-RPT4b proteins are observed in the same fractions as all other proteasome subunits, indicating that they assemble into the 26S complex.

Identification of plant proteasome assembly chaperones Affinity purification of Arabidopsis 26S proteasomes via the CP or RP
Using these FLAG-RPT4a and FLAG-RPT4b lines together with the PAG1-FLAG line described previously (29), we exploited a rapid affinity method to purify proteasomes from clarified seedling extracts using a single anti-FLAG chromatography step followed by elution with the FLAG peptide. As shown in Fig. 3A, isolations performed in the presence of ATP enabled strong enrichment of 26S particles containing both CP and RP sub-complexes, whereas those undertaken in the absence of ATP enabled strong enrichment of free CP or RP sub-complexes alone, with the characteristic array of lower molecular mass species (20 -37 kDa) for the CP and higher molecular mass species  for the RP being visible. The identities of these complexes were then confirmed by immunoblot analyses of the samples with antibodies against subunits from the CP (PAC1 (␣ 3 ), PAG1 (␣ 7 ), PBA1 (␤ 1 ), and PBF1 (␤ 6 )), the RP base (RPN1, RPT2, and RPT4), or the RP lid (RPN3, RPN5, and RPN12a; Fig. 3B). As expected, the FLAG-RPT4a/b isoforms co-purified with the complete 26S complex with ATP and with the RP alone without ATP, whereas the PAG1-FLAG protein co-purified with the complete 26S complex with ATP and with the CP alone without ATP (Fig. 3B).
To define the composition of 26S proteasomes purified via FLAG-RPT4a, FLAG-RPT4b, or PAG1-FLAG, we subjected trypsinized preparations isolated with or without ATP to deep protein sequencing by reverse-phase liquid chromatographic separation of the peptides followed by tandem MS (LC-MS/ MS). The relative levels of each protein were then determined by label-free quantification (LFQ) based on the MS1 precursor ion intensities (36). These analyses resulted in the unequivocal identification of all 14 subunits of the CP, all six RPT ring subunits, and 12 out of 13 RPN subunits (the exception being RPN13), including SEM1/RPN15 that had previously escaped detection (Fig. 4A) (29). In most cases (the exceptions being PAC2 (␣ 3 ), RPT1b, and RPN12b), polypeptides derived from both paralogs encoded within the Arabidopsis genome were detected. Previous studies suggested that these three predicted, but not detected, isoforms are encoded by pseudogenes (29,37). When the LFQ values for each protein were compared among the preparations generated with FLAG-RPT4a, FLAG-RPT4b, or PAG1-FLAG without ATP, an obvious enrichment was seen for CP subunits when using the PAG1-FLAG pag1-1 line, and for the RP subunits when using the FLAG-RPT4a rpt4a-1 or FLAG-RPT4b rpt4b-2 lines (Fig. 4A). In fact, estimates based on LFQ values revealed a 23-fold enrichment for the CP over the RP when purified via PAG1-FLAG, and a 3.4-fold enrichment of the RP over the CP when purified via FLAG-RPT4a/b (Fig. 5).

Arabidopsis 26S proteasomes do not display isoform specificity with respect to RPT4
Given the possibility that Arabidopsis and other plants use diversity among paralogous genes to generate distinct proteasome subtypes with unique specificities (29,31,(38)(39)(40), much like the exchange of mammalian ␤ 1 , ␤ 2 , and ␤ 5 subunit isoforms generate immuno-and thymo-proteasomes (12), we tested whether the RPT4a and RPT4b isoforms would selectively associate with isoforms from other proteasome subunits. In this case, we compared statistically the abundance of other isoforms in preparations obtained with FLAG-RPT4a or FLAG-RPT4b by volcano plots that displayed both fold change and p values of significance between the two samples (Fig. 4B). Based on the kernel density estimations (KDE) of their LFQ values, both samples displayed a strong overlap when comparing subunits isolated with either the entire 26S particle or with just the RP specifically (i.e. with and without ATP).
When comparing subunit isoforms individually, RPT4a and RPT4b were only modestly enriched relative to other subunits in the FLAG-RPT4a rpt4a-1 and FLAG-RPT4b rpt4b-2 samples obtained with ATP, respectively. Given that doubly-capped 26S proteasomes contain two RPs, each of which houses a single RPT4 polypeptide, this modest enrichment implied that 26S particles assemble with a mix of both RPT4a and RPT4b (Fig.  4B). Enrichment of other proteasome subunits or isoforms was A, affinity purification of 26S proteasomes from Arabidopsis seedlings expressing either PAG1-FLAG, FLAG-RPT4a, or FLAG-RPT4b in their corresponding pag1-1, rpt4a-1, or rpt4b-2 mutant backgrounds. Total protein extracts from 10-day-old seedlings were incubated with anti-FLAG beads in the presence or absence of 20 mM ATP and washed, and bound proteins were eluted with the FLAG peptide. Preparations were separated by SDS-PAGE and then stained for protein with silver. The position of the alternative capping particle PA200 is indicated by an arrowhead, and the positions of other CP and RP subunits are indicated by the brackets. B, proteasomes affinity-purified as in A were analyzed by SDS-PAGE followed by immunoblot with antibodies against the indicated proteasome CP and RP subunits, and the FLAG epitope. Open and closed arrowheads locate FLAG-tagged and endogenous (i.e. nontagged) proteasome subunits, respectively.

Identification of plant proteasome assembly chaperones
also not detected. However, when the RP was specifically enriched in samples obtained without ATP, and thus should contain only a single RPT4, the RPT4a and RPT4b polypeptides showed significant enrichment for their respective FLAGtagged version (Fig. 4B). Despite this, no other subunits, or subunit paralogs, were similarly enriched, indicating that RPT4a and RPT4b are not deliberately paired with other proteasome subunits or their isoforms. Based on this result with RPT4a/b, we found no evidence for the assembly of specific proteasomes types based on isoform preferences. The likely scenario is that Arabidopsis randomly assembles 26S proteasomes from the available pool of subunit isoforms in each cell, although a final conclusion will require similar analyses with CP subunits, especially with respect to the PBB1/2 (␤ 2 ) and PBE1/2 (␤ 5 ) paralogs.

Identification of a suite of CP-and RP-specific proteasomeinteracting proteins
Because our one-step protocol allowed rapid enrichment of 26S proteasomes from Arabidopsis tissues under nondenaturing conditions and without the need for stringent wash and elution steps, a number of proteasome-associated proteins remained bound and were then identified by LC-MS/MS (false discovery rate (FDR) Ͻ0.01; Table 1; Figs. 4C, and 5). Of particular note were proteins associated with proteasome sub-complexes purified in the absence of ATP, which could represent CP-or RP-specific interac- from three independent biological replicates. B, volcano plots comparing proteasomes affinity-purified via the FLAG-RPT4a or FLAG-RPT4b subunits in the presence (left) or absence (right) of ATP. The dashed lines demarcate proteins that were differentially represented in the enriched samples relative to their control, as judged by a significant analysis of microarray (SAM) test using an FDR of 0.01. Two-dimensional KDEs are shown for CP subunits (red) and RP subunits (blue). Individual CP and RP subunits are indicated as red and blue squares, respectively. Other proteins that did not reach SAM significance are shown as gray circles. The distribution of CP and RP subunits is shown below each volcano plot. The x axis shows the difference in CP (red) or RP (blue) protein levels between samples affinity-purified via FLAG-RPT4a or FLAG-RPT4b, with the area of each density defined as 1. Solid vertical lines represent the median values of either the CP (blue) or RP (red) subunits. Cryo-electron microscopic structures of the proteasome sub-complexes enriched in each sample are shown (adapted from Ref. 32), with the CP in red, the RP base in blue, and the RP lid in gold. C, LFQ-MS analysis of proteasome-associated proteins that were enriched via the PAG1-FLAG, FLAG-RPT4a, or FLAG-RPT4b subunits in the presence or absence of ATP. Values represent the mean (ϮS.D.) from three independent biological replicates. A zero value of 15 was imputed to better visualize differences. The color code for the bars is the same as in A.

Identification of plant proteasome assembly chaperones
tors. Such specific associations were immediately evident in volcano plots of the preparations (performed with or without ATP), with the interacting proteins showing clear overlap based on KDEs associated with either the CP or RP (Fig. 5).
For the CP samples obtained using PAG1-FLAG, these interacting proteins likely included homologs of the proteasome maturation factor UMP1 (22), the extrinsic ubiquitin-binding receptor DSK2 (41,42), plus REDUCED CHLOROPLAST COVERAGE 1 (REC1 (43)), and an F-box protein of unknown function (Table 1). Importantly, four of the most highly enriched proteins were identified by position-specific iterative (PSI) and reciprocal Basic Local Alignment Search Tool (BLAST) searches as possible orthologs of the previously described yeast and mammalian CP assembly chaperones Pba1-4 and PAC1-4, respectively (13,14), despite low sequence identity (9 -30%; see Fig. 6A, and Figs. S2-S6). As these three-letter annotations were already in use (29), we renamed these Arabidopsis proteins as PROTEASOME BIOGENESIS-ASSOCIATED CHAPERONE (PBAC)1-4. PBAC1 and PBAC2 were previously identified by PSI-BLAST searches with their yeast counterparts (16,17,29), and their identities were also supported by the presence of C-terminal HbYX or related HbF motifs, respectively, which are universally present in the yeast, archaeal, and mammalian members of these two chaperone families (Figs. S2A and S3A). As described above, HbYX motifs are employed by assembly chaperones to bind and stabilize association of the ␣-subunits during initial ␣-ring formation, as well as being used by other activators (including the RP itself) to mediate their binding to the CP and control ␣-ring gate opening during substrate degradation (11). An additional PBAC-type protein with a HbYX motif was also found associated with the Arabidopsis CP (designated PBAC5), which likely  Fig. 3A, and subjected to tandem MS. The dashed lines demarcate proteins that were differentially represented in the enriched samples relative to their control, as judged by a SAM test using an FDR of 0.01. Two-dimensional KDEs are shown for CP subunits (red) and RP subunits (blue). Individual CP and RP subunits are indicated as red and blue squares, respectively. Proteins likely involved in CP assembly are shown as green squares, whereas proteins likely involved in RP assembly are shown as yellow squares. Proteasome-associated proteins with unknown or possible regulatory functions are shown as gray and black squares, respectively. Other proteins that did not reach SAM significance are shown as gray circles. The distribution of CP and RP subunits is shown below each volcano plot. The x axis shows the difference in CP (red) or RP (blue) protein levels between affinity-purified samples and their respective controls, with the area of each density defined as 1. Solid vertical lines represent the median values of either the CP (blue) or RP (red) subunits. Fold changes (FC) for CP abundance compared with the RP (for PAG1-FLAG samples), or for RP abundance compared with the CP (for FLAG-RPT4a/b samples) were calculated based on the median of each distribution. p values for median and KS tests are shown. Cryo-electron microscopic structures of the proteasome sub-complexes likely to be enriched in each sample are shown (adapted from Ref. 32), with the CP in red, the RP base in blue, and the RP lid in gold.

Identification of plant proteasome assembly chaperones
represents a fifth member of the PBAC superfamily; its characterization is currently in progress. 6 An additional list of interacting proteins was obtained for the RP enriched without ATP (Table 1; Figs. 4C and 5). Notably, three of the top interactors identified in both FLAG-RPT4a and FLAG-RPT4b samples were identified by PSI and reciprocal BLAST searches as possible orthologs of the previously described yeast and mammalian RP assembly chaperones NAS2/p27, NAS6/p28, and HSM3/S5b (13,14), again despite low sequence identity (10 -36%; see Fig. 6A and Figs. S7-S9). As highlighted in Fig. 6A, all the possible CP and RP chaperones have relatively high sequence identities within the plant kingdom, but this homology dropped considerably when compared with their yeast and human relatives. Additional high-confidence interactors that co-purified with both the CP and RP included homologs of the known alternative proteasome cap PA200/Blm10 (29), the proteasome assembly factor ECM29 (26,28), and the proteasome regulator PTRE1 (Table  1; Fig. 5; Fig. S10) (30). Also of interest were the AGAMOUS-LIKE 90 (AGL90) transcription factor and the MULTI-DRUG RESIS-TANCE-ASSOCIATED PROTEIN 8 (MRP8), although the functional significance of these two interactions is currently unknown.
We considered it possible that some proteasome interactors could be direct targets being degraded by the complex. However, few if any matched known targets or the deep lists of ubiquitylation substrates from Arabidopsis (44,45). In hindsight, this absence was likely given the speed that proteasomes can process substrates (46) and the fact that the proteasomes were not purified in the presence of inhibitors that would block continued target hydrolysis after tissue extraction.

Proteasome assembly chaperones have distinct expression patterns in Arabidopsis
Under the assumption that proteasome assembly chaperones should be coordinately expressed to meet proteolytic demand, we attempted to provide support that the Arabidopsis collection (PBAC1-4, UMP1, NAS2, NAS6, HSM3, and ECM29) is indeed involved in proteasome assembly. However, examination of their expression patterns in 79 Arabidopsis tissues available within the Transcriptome Variation Analysis database (www.travadb.org) 7 (47) failed to Identification of plant proteasome assembly chaperones find a common theme, even among factors that should work together (e.g. PBAC1-4; Fig. 6B). However, when seedlings were activated to express 26S proteasomes (and thus their assembly chaperones) by treatment with the protease inhibitor MG132 (48) or by the introduction of subunit mutations that compromise proteasome activity (33,37), several of these loci were found to be up-regulated along with the rest of the proteasome-stress regulon (Fig. 6C). The extent of the up-regulation varied; whereas the transcript abundance for PBAC1, PBAC2, UMP1a, NAS6, HSM3, and ECM29 increased between 4-and 8-fold after 3 h of MG132 treatment, those for PBAC3, PBAC4, and NAS2 were only modestly increased even after a 24-h exposure (on average around 1.4-fold).

Assembly chaperones associate with proteasome assembly intermediates
As a second strategy to connect the proposed chaperones to proteasome assembly, we tested whether they could be found linked to the various intermediates that should arise during holo-26S proteasome construction in planta (35,49). Here, the accumulation of these complexes was enhanced by pre-treating Arabidopsis seedlings with MG132 to encourage new proteasome synthesis (Fig. 6C) (48), and the assembly intermediates, along with the CP, RP, and 26S particles, were then affinityenriched via the PAG1-FLAG tag in the presence of ATP. Although the profile of proteasome subunits was not dramatically altered by this treatment (Fig. 7A), the effects of MG132 on both proteasome assembly and capacity were evident by the increased ubiquitylation of the complex (35), increased association of the PA200/Blm10 regulator (29), and by the appearance of novel proteasome sub-particles in addition to the CP, RP, and holo-26S complex after separation by native PAGE (Fig. 7, A-C).
To define the protein composition of each species in the native gels, the corresponding bands were excised and in-gel digested with trypsin, followed by LC-MS/MS analysis (Fig.  7D). When the protein composition of each band was quantified by distributed Normalized Spectral Abundance Factor (dNSAF) values (50,51), several assembly intermediates were identified, along with stable CP and RP sub-complexes, that were consistent with PBAC1-4 and UMP1 being involved in CP assembly (Fig. 7, C and D).
The smear of species detected at the bottom of the native gels, especially after MG132 treatment of the seedlings, was enriched in CP ␣-subunits, with low levels of ␤-subunits, together with the proposed CP assembly chaperones PBAC1-4, suggesting that these species mostly comprised partially or fully assembled ␣-rings (17). Indeed, PBAC3 and PBAC4 were found only in this region of the gels. In agreement with this smear containing partially assembled CPs, a low level of UMP1 was also detected in this region. The prominent band #1 labeled in Fig. 7C contained only CP subunits (and not PBAC1-4), indicating that this species represented fully assembled CP, whereas the minor band #2 seen above upon MG132 treatment contained CP plus PBAC1-2, suggesting that it represented nearly assembled 15S half-barrels still associated with CP assembly chaperones (Fig. 7, C and D). Band #3 contained CP subunits together with PA200/Blm10 Figure 6. Arabidopsis proteasome assembly chaperones have weak sequence homology and distinct expression patterns. A, heat map displaying the percentage amino acid sequence identity of nine predicted Arabidopsis proteasome assembly chaperones identified in Fig. 4, compared with those for orthologs from poplar, maize, moss, yeast, and humans. The percentage sequence identity between each chaperone from the above species was calculated in Clustal Omega. B, heat map displaying the relative transcript abundance for the predicted Arabidopsis proteasome assembly chaperones shown in A, plus UMP1c, in 32 manually curated tissues. Expression data for UMP1b was not available. mRNA levels were obtained from the Transcriptome Variation Analysis database and hierarchically clustered by both tissue distribution and expression patterns. All read counts for each gene were normalized by the "median-of-ratios" method, with the maximum value set at 1. C, heat map displaying the expression levels of transcripts encoding the predicted Arabidopsis proteasome assembly chaperones in Arabidopsis seedlings treated with MG132, or in the Arabidopsis rpn10-1, or rpn12a-1 mutants with impaired proteasome function. Data were obtained from a previously published RNA-seq dataset (GSE81668 (45)).

Identification of plant proteasome assembly chaperones
but little PBAC1-2, suggesting it contained CPs likely capped at one end by this CP regulator (Fig. 7, C and D). In MG132treated samples, bands #4 and #5 appeared, whose compositions were consistent with them representing the CP capped with PA200/Blm10 and bound by the PBAC1-2 assembly chaperones, and the CP capped at both ends by PA200/Blm10, respectively. Bands #1-5 also contained ubiquitin, especially in samples from MG132-treated seedlings, which could reflect ubiquitylated CP or its intermediates now targeted for turnover by autophagy (35).
Moving further up the native gels, bands #6 and #7 were mostly enriched for RP subunits, with low levels of CP subunits, and thus represented mostly free RP sub-complexes (Fig. 7, C and D). As these RP species were isolated solely based on their association with CP harboring PAG1-FLAG, they should represent RPs that dissociated from the CP after isolation. Finally, bands #8, #9, and #10 appeared to contain equimolar amounts of subunits for both the CP and RP sub-complexes and likely represented the fully assembled holo-26S particle containing either one or two RPs. Interestingly, only band #10 was found to contain the Arabidopsis ortholog of ECM29, albeit at low levels ( Fig. 7, C and D), as might be expected if ECM29 helps tether the RP to the CP (26,27). The possible RP chaperones NAS2, NAS6, and HSM3 were not detected during this analysis, presumably because the proteasomes were enriched via the CP subunit PAG1 and thus should contain only fully assembled RP species.

PBAC3 and PBAC4 can rescue yeast mutants missing their orthologs
As a third strategy to confirm that these chaperones are involved in proteasome assembly, we attempted to rescue yeast mutants missing Pba3 and Pba4 with their likely Arabidopsis orthologs, PBAC3 and PBAC4, respectively. As shown in Fig.  8A, yeast ⌬pba3 and ⌬pba4 strains are highly sensitive to proteotoxic stress, which can be seen by suppressed colony growth at low micromolar concentrations of the amino acid analog proteasomes upon inhibition. Total protein extracts from 10-day-old WT or PAG1-FLAG pag1-1 seedlings treated with or without 50 M MG132 were incubated with anti-FLAG beads, the beads were washed, and bound proteins were eluted with FLAG peptide. Preparations were separated by SDS-PAGE and then stained for protein with silver (left) or immunoblotted with anti-ubiquitin (Ub) antibodies (right). The position of the alternative capping particle PA200 is indicated by an arrowhead, and the positions of other CP and RP subunits are indicated by the brackets. B and C, separation of the proteasome preparations from A by native PAGE for 16 h (B) or 36 h (C). The gels were then stained for total protein with silver. Migration positions of various complexes, including free CP, CP capped with PA200, free RP, singly-and doubly-capped holo-26S proteasomes, CP complexes containing predicted assembly chaperones, and free ␣-rings are indicated based on tandem MS analysis of the samples, as shown in D. The dashed boxes represent gel sections that were excised, trypsinized, and subjected to tandem MS analysis in D, The asterisk represents an unknown assembly intermediate. D, LFQ-MS analysis of the protein compositions of the various proteasome species affinity-purified following inhibition. Shown is a heat map displaying dNSAF values for specific proteins in the gel slices shown in B and C, which were calculated using Morpheus Spectral Counter and then hierarchically clustered by Pearson's correlation. Smear represents the diffuse species detected at the bottom of the native gel shown in B.

Identification of plant proteasome assembly chaperones
canavanine. Both strains were equally hypersensitive, consistent with Pba3 and Pba4 working as heterodimers (20,21). Strikingly, strong rescue of this growth defect on 5 M canavanine was evident upon introduction of tagged versions of both Arabidopsis PBAC3 and PBAC4. Like yeast Pba3, HA-tagged PBAC3 rescued the defect in ⌬pba3 cells but not in ⌬pba4 cells, whereas HA-tagged PBAC4 rescued the defect in ⌬pba4 cells but not in ⌬pba3 cells (Fig. 8A). This lack of cross-complementation, plus the failure of ⌬pba3 ⌬pba4 cells to be rescued by the two chaperones individually, further supported the non-redundant functions of PBAC3 and PBAC4 and implied that these Arabidopsis proteins also work as heterodimers.

UMP1 is a CP maturation factor essential in Arabidopsis
In a parallel fashion, we attempted to demonstrate that Arabidopsis UMP1 is orthologous to yeast Ump1, despite low sequence identity (18, 16, and 13% for Arabidopsis UMP1a, UMP1b, and UMP1c, respectively), with our first approach being yeast complementation assays as above. As shown in Fig.  8B, growth of yeast ⌬ump1 cells is similarly hypersensitive to 5 M canavanine and also to heat stress at 37°C (22). Whereas HA-tagged yeast Ump1 rescued both of these growth defects, we were surprised to find that the three putative Arabidopsis orthologs all failed (Fig. 8B), even though each was adequately expressed in the ⌬ump1 background, based on immunoblots of total cell extracts with anti-HA antibodies (Fig. 8C). Complementation of the canavanine-sensitivity phenotype was also not observed when combinations of two, or even all three, Arabidopsis UMP1 isoforms were expressed together, indicating that they do not function as heterodimers or heterotrimers (Fig. 8B).
As an alternative, we tested whether Arabidopsis ump1 mutants harbor a similar hypersensitivity to proteotoxic stress induced by agents such as canavanine and MG132. As mentioned above, Arabidopsis potentially encodes three UMP1 paralogs, two of which (UMP1a (At1g67250) and UMP1b (At5g38650)) align reasonably well to yeast Ump1 and its mammalian counterpart POMP (Fig. 9, A and B, and Fig. S6). Only UMP1a is transcriptionally up-regulated by conditions that impair proteasome function, implying that it could be the dominant isoform (Fig. 6C). The third paralog, UMP1c (At1g62920), shares homology within the central core region but is predicted to be missing 58 amino acids at the N terminus and to contain an additional 152 amino acids at the C terminus, neither of which are well-conserved among UMP1 proteins in general or among plant UMP1 proteins in particular (Figs. 9, A  and B, and Fig. S6).
From a search of the Arabidopsis T-DNA insertion collections, we identified potentially useful mutants in all three Arabidopsis UMP1 genes, which were confirmed by genomic DNA sequencing around the insertion sites (Fig. 9, A and B). Transcript analyses by RT-PCR failed to detect corresponding fulllength mRNAs in several of the alleles when homozygous (ump1a-2, ump1a-3, ump1b-2, ump1c-1, ump1c-2, and  ump1c-5; Fig. 9C). Given the positions of the T-DNAs, the

Identification of plant proteasome assembly chaperones
ump1a-2, ump1b-2, and ump1c-2 alleles were studied in detail. For the ump1a-2 allele, no partial transcripts were amplified, whereas for the ump1b-2 and ump1c-2 alleles, only transcripts generated upstream of the insertion site could be detected, suggesting that these mutants represent strong, if not null, alleles (Fig. 9C). Under standard growth conditions with a long-day photoperiod, plants homozygous for each mutant appeared phenotypically normal and produced viable seeds, indicating that none of the three UMP1 loci are essential to Arabidopsis growth and development by themselves.
In an attempt to obtain Arabidopsis lines missing all three UMP1 isoforms, we began generating higher-order mutant combinations by introgressing the ump1a-2, ump1b-2, and ump1c-2 lines. Although ump1a ump1c and ump1b ump1c double-homozygous progeny could be easily identified with near normal Mendelian segregation ratios from a self-cross of double-heterozygous parents, we surprisingly failed to find double-homozygous ump1a ump1b progeny (Fig. 9, E-G), suggesting that, unlike in yeast (22), UMP1 is essential to Arabidopsis. In addition, ump1a ump1b plants that were

Identification of plant proteasome assembly chaperones
heterozygous for one allele and homozygous for the other were identified by at a much lower frequency than expected ( Fig. 9E; 2 test, p ϭ 6.33 ϫ 10 Ϫ6 ). Visual inspection of siliques from a selfed ump1a-2/ϩ ump1b-2/ϩ doubleheterozygous plant revealed the presence of numerous aborted seeds (Fig. 9D), indicating that UMP1a and UMP1b act redundantly to ensure proper seed development.
The fact that ump1a ump1b double mutants were not viable, but ump1a ump1c and ump1b ump1c double mutants were easily identified, suggested that UMP1c does not play an important role in plant proteasome assembly. In support, our LC-MS/MS analyses failed to provide experimental evidence that UMP1c associates with proteasomes, even from a directed search of tryptic peptides from its extended C-terminal domain that could differentiate it from UMP1a/b (Figs. 4C and 5). Consequently, we concluded that the UMP1c protein likely does not accumulate, despite evidence that the corresponding UMP1c locus is expressed (Fig. 6B).

Arabidopsis UMP1 protects against proteotoxic stress by encouraging CP assembly
To test whether Arabidopsis ump1 mutants, like their yeast counterparts (22), are hypersensitive to proteotoxic stress, we measured the growth response of single homozygous ump1a, ump1b, or ump1c seedlings when exposed to various concentrations of canavanine and MG132, using the response of double mutant seedlings missing the NAC53 and NAC78 transcription factors (which are essential for activating the proteasome-stress regulon (48)) as a positive control. As shown in Fig. 10, A and B, WT seedlings grew reasonably well on sub-lethal doses of MG132 or canavanine beginning at germination, and became impacted only at higher doses (50 M MG132 and 25 M canavanine). By contrast, growth of the nac53-1 nac78-1 seedlings was strongly inhibited by even low concentrations of the drugs.
In agreement with UMP1c being a possible pseudogene, three different mutant alleles (ump1c-1, ump1c-2, and ump1c-5) were indistinguishable from WT in their tolerance to long-term MG132 or canavanine exposure (Fig. 10, A and B). However, alleles impacting either UMP1a or UMP1b showed a strong hypersensitivity to both treatments. The growth inhibition of both UMP1a alleles (ump1a-1 and ump1a-2) was noticeably stronger than that for UMP1b (ump1b-2), suggesting again that UMP1a is the dominant locus (Fig. 10, A and B). This dominance could be caused by its higher overall expression level; for example, 90, 55, and 7 ESTs were found for UMP1a, UMP1b, and UMP1c, respectively, in the Arabidopsis Information Resource database (version 10.1).
Recently, an Arabidopsis mutant in the proteasome subunit PBE1 (␤ 5 ) was shown to have impaired CP assembly, resulting in a hypersensitivity to salt and drought stress, presumably due to insufficient levels of the CP or the entire 26S holocomplex (49). We therefore tested whether our panel of ump1 single mutants was similarly sensitive to such treatments. While we again found that the ump1c mutants grew equivalent to WT, the growth of both the ump1a and ump1b mutants was strongly impaired on medium containing NaCl, mannitol, or the drought-responsive hormone abscisic acid (ABA) at concentra-tions known to only mildly suppress WT Arabidopsis growth (Fig. 11, A and B). Notably, the consequences were similar but less pronounced than for the null pbe1-2 mutant inactivating the dominant locus encoding this catalytically active CP subunit (49). The ump1a and ump1b seedlings thus display phenotypic defects similar to another proteasome mutant with dampened CP capacity.
To definitively assess the importance of UMP1 to proteasome assembly, we monitored the integrity of the complex by glycerol gradient fractionation of total-cell extracts from WT and ump1a-2 seedlings with or without a 16-h pretreatment with 50 M MG132. Consistent with prior results in Fig. 2, a single proteasome species was detected in WT seedlings by immunoblot analysis of the fractions with a panel of proteasome subunit antibodies, which likely represented the 26S holo-complex (Fig. 10C). Upon exposure to MG132, substantial amounts of free RP and CP subunits accumulated at the top of the gradient, as might be expected based on activation of the proteasome stress regulon by this inhibitor in an effort to increase 26S proteasome supply (48). In contrast, multiple additional species were seen within the gradient from untreated ump1a-2 seedling extracts besides free subunits and the 26S complex (Fig. 10C). Based on sedimentation position, these likely reflected free RP, free CP ␤-subunits, and what appeared to be CP ␣-subunits organized into ␣-ring intermediates.
When we combined the ump1a-2 mutation with MG132 treatment, a host of proteasome species appeared that were consistent with up-regulated proteasome subunit synthesis combined with impaired CP (and thus holo-26S particle) assembly. Some of these intermediate species could also represent the CP singly-or doubly-capped with PA200 ( Fig. 7 C) (29,35). Remarkably, little of the fully-assembled holo-26S proteasome complex was evident relative to the sub-complexes, thus explaining why ump1a-2 seedlings grew so poorly when exposed long-term to MG132 (Fig. 10, A and B). Taken together, our observations that mutants impacting Arabidopsis UMP1 accumulate a variety of intermediate proteasome complexes, generate the same hypersensitivity to proteotoxic stress as do yeast ⌬ump1 strains, and display a similar hypersensitivity to salt, drought, and ABA stress as an Arabidopsis CP mutant with known assembly defects, provide strong support for UMP1 being crucial for Arabidopsis CP assembly.

Discussion
The 26S proteasome plays a critical role in plants by degrading key regulatory proteins and eliminating mis-folded proteins that might otherwise become cytotoxic. Given the sophisticated architecture of this multisubunit particle, it is unsurprising that numerous intrinsic and extrinsic factors combine to promote rapid and faithful assembly. However, although studies on proteasome assembly in mammals and yeast began almost 2 decades ago (13,14), understanding of this pathway in plants has remained sparse. Here, we described affinity methods to rapidly isolate the 26S particle from Arabidopsis based on the RP which, when combined with a prior method to isolate the CP (29), not only validated the composition of the plant particle but also identified a suite of CP-and RP-specific inter-

Identification of plant proteasome assembly chaperones
acting proteins that presumably assist in particle assembly, activity, and/or regulation.
MS analyses of 26S proteasomes purified via PAG1, RPT4a, or RPT4b invariably identified the complete collection of CP and RP subunits, including all encoded isoforms except for PAC2, RPT1b, and RPN12b, which are predicted to be pseudogenes (29). The use of the RPT4a and RPT4b paralogs for affinity enrichment also allowed us to test the hypothesis that Arabidopsis assembles distinct proteasome isotypes based on the selective incorporation of specific subunits or their isoforms. No differences in subunit or paralog compositions could be found between proteasomes purified via RPT4a or RPT4b, suggesting that, at least with respect to RPT4, proteasome subunit isoforms are assembled randomly. However, evidence has recently emerged that the expression of certain plant proteasome subunits increases in response to proteotoxic or salt stress, or upon elicitation of defense responses, suggesting that unique subunit expression patterns, rather than specific isoform incorporation, could play a role in diversifying proteasome functions in plants (48,49,52,53). Whether these changes result in altered substrate-processing properties, as observed with the immuno-or thymo-proteasomes found in humans, remains to be determined.
Purification of proteasomes in the absence of ATP, which allows for CP-RP dissociation, permitted specific enrichment of the CP and RP sub-complexes, thus enabling identification of numerous CP-and RP-specific interacting proteins. Importantly, several of these were homologous to known yeast and mammalian proteasome assembly chaperones. Included were relatives of yeast Pba1-4, Ump1, Nas2, Nas6, Hsm3, and Ecm29, along with the CP regulator PA200/Blm10. Amino acid sequence homology was surprisingly low among the chaperones, especially when comparing the plant sequences to their animal and fungal brethren (see Figs. S1-S10). For example, there was never greater than 18% amino acid identity between a plant and non-plant PBAC1 ortholog, and in the majority of cases the identities were less than 12%. Even within the plant kingdom, PBAC1 had over 70% identity among the four monocot species analyzed (Fig. S2C) but dropped to 45% or less when compared with those in pineapple, Selaginella, C, ump1a-2 mutant accumulates proteasome assembly intermediates. Total protein extracts from 10-day-old WT or ump1a-2 seedlings treated with or without 50 M MG132 for 16 h were subjected to glycerol gradient fractionation, and samples from each fraction were analyzed by SDS-PAGE followed by immunoblot with antibodies against the indicated proteasome subunits. The location of each protein within the 26S complex is indicated in the right panel. The predicted positions of free proteasome subunits (free), assembly intermediates (interm.), free CP, free RP, and the holo-26S proteasome are indicated by the horizontal brackets. Numbers on the left represent molecular mass markers in kDa.
Such low identity among the chaperones implies that secondary structure or clusters of amino acids are more important than the overall primary sequence. Indeed, crystal structures of heterodimers formed by the yeast and human assembly chaperones Pba3-Pba4 and PAC3-PAC4, respectively, showed that the two complexes adopt essentially identical three-dimensional folds despite little sequence conservation (21, 54,55). Although individual residues were not well-conserved, structural features such as surface-charge complementarity were maintained. Especially notable was the presence of conserved HbYX-type motifs at the C terminus of Arabidopsis PBAC1 and PBAC2, which likely help these CP assembly chaperones dock with the ␣-subunit ring of the CP To help validate that the predicted chaperones are indeed involved in proteasome assembly, we examined their expres-sion patterns, tested for possible association with assembly intermediates of the CP, confirmed by complementation that some could replace their yeast relatives, and tested whether Arabidopsis mutants missing one of these chaperones might be hypersensitive to proteotoxic stress, as seen for comparable yeast mutants. Although the expression patterns for the collection showed little developmental co-ordination, co-regulated expression upon proteotoxic stress induced by MG132 or proteasome mutations was evident for most (PBAC1, PBAC2, UMP1a, NAS6, HSM3, ECM29 and, to lesser extents, PBAC3, PBAC4, and NAS2), thus including these loci in the proteasome-stress regulon that ensures an adequate supply of 26S particles (48,56). MS analysis of CP assembly intermediates induced by treating PAG1-FLAG seedlings with MG132 further connected PBAC1-4, UMP1, and ECM29 with assembly of the CP and association of the CP with the RP.
Complementation studies using Arabidopsis PBAC3 and PBAC4 to rescue the hypersensitivity of yeast strains missing Pba3 and Pba4 to proteotoxic stress definitively showed that these two are assembly chaperones. However, similar studies with combinations of any of the three Arabidopsis UMP1 paralogs failed, suggesting that these proteins have sufficiently diverged to prevent cross-complementation, as was also found to be the case with the mouse and human versions of Ump1

Identification of plant proteasome assembly chaperones
(known as POMP (57)). Alternatively, we generated a panel of Arabidopsis mutants missing UMP1a-c and showed that the lines missing UMP1a and UMP1b are hypersensitive to conditions that elicit proteotoxic stress, such as canavanine or MG132 treatments, as well as salt and drought stress, and displayed defects in proteasome assembly, much like their yeast Ump1 ortholog. Such assembly defects were particularly acute when combining the ump1a-2 mutation with MG132 exposure, with the Arabidopsis seedlings accumulating an array of sub-complexes and little of the 26S holo-complex. Surprisingly, although yeast strains missing Ump1 are viable, we failed to find Arabidopsis lines simultaneously missing the dominant isoforms UMP1a and UMP1b, indicating that this chaperone is essential in plants.
During the preparation of this manuscript, the maize mutation defective kernel 40 (dek40) was described to encode a possible ortholog of Arabidopsis PBAC4 (58). The dek40 seeds displayed abnormal development, while the shoots and roots showed slower growth after germination. Both overall proteasome activity and CP assembly were impaired in dek40 plants, which also hyper-accumulated ubiquitin conjugates presumably awaiting breakdown (58). Consistent with DEK40/PBAC4 working in CP assembly, its association with the maize ortholog of PBAC3 was also seen. However, despite defective proteasome assembly, the dek40 plants were still viable, suggesting that, unlike UMP1, the PBAC4 assembly chaperone is not essential in plants, as is also the case in yeast and mammals (16,19,20). Clearly, genetic analysis of the remaining plant chaperones is needed to determine which are essential for holo-26S proteasome assembly.
In addition to the chaperones described here, two other RP assembly chaperones have been identified in yeast, termed Rpn14 and Adc17 (13,59). However, no clear Arabidopsis relatives of Rpn14 or Adc17 were identified here by MS analysis of our RPT4-based affinity purifications, or by PSI-BLAST searches of the Arabidopsis genome using nonplant queries (Table 1; Fig. 5). Although some homologs of Adc17 were found in chlorophyte and streptophyte algal species (e.g. Chlamydomonas, Chlorella, Klebsormidium, Micromonas, and Volvox spp.), none were confirmed to exist in any higher plant genome. We did find one annotated sequence related to Adc17 from cork oak (Quercus suber; Uniprot accession number A0A2P4GUU7), but this sequence is remarkably similar to fungal Adc17 sequences, and thus most likely represents genomic contamination.
Besides assembly chaperones, we also found a variety of proteasome-associated factors that might help regulate its integrity and activity, including PTRE1, PA200/Blm10, and the ubiquitin-binding shuttle factor DSK2. PTRE1 and its animal relative PI31 are of particular interest given their purported role in modulating proteasome activity in vivo (5) and their strong enrichment in Arabidopsis proteasome preparations (this report). Deletion of PTRE1 in Arabidopsis generates an auxin hyposensitive phenotype, with elevated levels of the AUX/IAA family of auxin-response repressors, suggesting that it promotes auxin signaling by controlling UPS-mediated AUX/IAA protein turnover (30). Mutants eliminating PA200 do not display defects in phenotype, ubiquitin conjugate accumulation, proteasome activity, or sensitivity to proteasome inhibitors (29), but a role for PA200 in proteasome regulation is likely given its enhanced binding to the CP under conditions that induce proteotoxic stress (Fig. 7C) (29,35) and its observed role in chaperoning the CP into cytoplasmic proteasome storage granules during fixed-carbon starvation (60).
A number of plant proteins that are known orthologs of yeast or mammalian proteasome interactors were notably absent from our affinity purifications. These include the ubiquitinbinding shuttle factors RAD23 and DDI1 (42), the E3 ligase UPL7/HUL5 (28), and the deubiquitylating enzyme UBP6 (28, 61) that help control substrate degradation. It is possible that (i) these proteins interact with Arabidopsis proteasomes at wellbelow stoichiometric levels and thus evaded detection; (ii) the interactions are sensitive to even the mild wash and elution conditions employed here; (iii) the plant versions of these proteins do not interact with the proteasome under the conditions tested; and/or (iv) these proteins are not easily identified by MS. Clearly, additional studies will be required to understand why these proteins appear absent in Arabidopsis 26S proteasome preparations.
In summary, LC-MS/MS analysis of Arabidopsis proteasomes affinity-purified via either the CP or RP revealed a number of proteins interacting with a specific sub-complex, many of which are orthologs of known yeast and mammalian CP and RP assembly chaperones. These include PBAC1, PBAC2, PBAC3, and PBAC4 that likely promote CP assembly; UMP1 that is required for connecting CP half-barrels; NAS2, NAS6, and HSM3 that likely assist in RP assembly; and ECM29 that should provide a quality control checkpoint during CP-RP association. Subsequent mutational analysis of these chaperones should help confirm their roles in plant proteasome assembly and reveal how conserved the process is in plants compared with other organisms. For example, the existence of a possible fifth PBAC-type chaperone invites speculation of possible plant-specific proteasome assembly pathways. Using this list of chaperones, it should now be possible to define plant proteasome assembly through genetic and biochemical approaches in an effort to understand how it might be manipulated for improved crop yield, stress protection, and nutrition.

Plant materials and growth conditions
All A. thaliana plants were derived from the Columbia-0 (Col-0) ecotype, except the ump1c-4 allele, which was from the Col-3 ecotype. Details of all T-DNA insertion mutants and stable transgenic lines are provided in Tables S1 and S2. All insertion mutants were confirmed by genomic PCR using 5Ј and 3Ј gene-specific primers (LP and RP, respectively) in conjunction with an appropriate T-DNA left border-specific primer. Details of all oligonucleotide primers are provided in Table S3. Exact T-DNA insertion positions were determined by direct sequencing of appropriate PCR products. Before analysis, all mutants were backcrossed three times to the Col-0 parent and then selfed to obtain homozygous progeny, using growth on medium containing 50 g/ml kanamycin and/or 10 g/ml BASTA for Identification of plant proteasome assembly chaperones selection. Sterilized seeds (obtained via vapor-phase or liquidphase sterilization) were vernalized at 4°C for 3-4 days and typically germinated on solid GM medium (3.2 g/liter Gamborg's B5 basal medium with minimal organics, 1% (w/v) sucrose, 0.05% (w/v) MES (pH 5.7), 0.7% (w/v) agar) at 21-23°C under a long-day photoperiod (16-h light/8-h darkness), with a light intensity of 75-100 mol/m 2 /s and a relative humidity of 40 -50%. Where indicated, agar plates were supplemented with the indicated concentrations of MG132 ((N-benzyloxycarbonyl)-leucinyl-leucinyl-leucinal), canavanine, NaCl, mannitol, or ABA. After 2-3 weeks, seedlings were transferred to soil (mixed in a 1:1 ratio with organic Coco Coir planting mixture, supplemented before use with 2 g/liter Peters 20-20-20 fertilizer, 80 mg/liter Ca(NO 3 ) 2 , and 80 mg/liter MgSO 4 ), and grown at 21-23°C under a long-day photoperiod.
To generate transgenic plants expressing FLAG-tagged RPT4a, the genomic region of RPT4a encompassing the fullcoding sequence plus 600-bp upstream of the ATG start codon, which includes the 5Ј-and 3Ј-untranslated regions (UTRs), was PCR-amplified from Col-0 genomic DNA and recombined into pDONR221 via the Gateway BP clonase II reaction (Thermo Fisher Scientific). Codons for the 2ϫFLAG tag (DYKDDDDK-DYKDDDDK) were then inserted by successive rounds of mutagenesis using the QuikChange II site-directed mutagenesis kit (Agilent Technologies). The sequence-confirmed RPT4a::2ϫFLAG-RPT4a clone was then recombined into the pMDC123 plant transformation vector (62) via the Gateway LR clonase II reaction (Thermo Fisher Scientific). To generate plants expressing FLAG-tagged RPT4b, the genomic region of RPT4b encompassing the full-coding sequence plus 2,000-bp upstream of the ATG start codon, which again includes the 5Јand 3Ј-UTRs, was PCR-amplified from Col-0 genomic DNA using primer pairs encoding either the 2ϫFLAG tag plus NcoI and SmaI restriction enzyme sites, or the PstI and NcoI restriction enzyme sites, for the coding sequence and upstream region, respectively. The resulting PCR products and the recipient pCAMBIA3301 plant transformation vector were digested with FastDigest NcoI, SmaI, and/or PstI (Thermo Fisher Scientific), and ligated together using T4 DNA ligase (New England Biolabs) to generate the RPT4b::2ϫFLAG-RPT4b clone.
Sequence-confirmed FLAG-RPT4a and FLAG-RPT4b plasmids were introduced into Agrobacterium tumefaciens strain GV3101 and transformed into homozygous rpt4a-1 or rpt4b-2 plants, respectively, by the Agrobacterium-mediated floral dip method (63). BASTA-resistant F 1 plants were selected and, after a self-cross, double-homozygous plants were identified in the F 2 generation by PCR genotyping, immunoblot analysis (see below), and segregation of the F 3 generation on BASTA-containing medium.
For immunoblot analyses, proteins separated by SDS-PAGE were electrophoretically transferred onto Immobilon-P polyvinylidene difluoride membranes (Millipore) for 16 h at 80 mA, and the membrane was blocked for at least 60 min with a 10% (w/v) non-fat dry milk solution in 1ϫ PBS (137 mM NaCl, 2.7 mM KCl, 10 mM Na 2 HPO 4 , 1.8 mM KH 2 PO 4 ). All incubations were performed at room temperature. The membrane was incubated with primary antibody solution (in 1% (w/v) non-fat dry milk solution in PBS) for 60 min, before being washed once with PBS, once with PBST (PBS containing 0.1% (v/v) Triton X-100), and once with PBS for 10 min each. The membrane was re-blocked with 10% (w/v) non-fat dry milk solution in PBS for 30 min, incubated for 60 min with secondary antibody solution (in 1% (w/v) non-fat dry milk solution in PBS), and then washed again as above. Blots were developed using the SuperSignal West Pico Plus Chemiluminescent Substrate or the SuperSignal West Femto Maximum Sensitivity Substrate (both from Thermo Fisher Scientific).
To test antigenicity of the anti-RPT4a antibodies, the coding sequences of RPT4a and RPT4b were PCR-amplified from Col-0 cDNA (generated as described below) using primers

Identification of plant proteasome assembly chaperones
encoding an NheI restriction enzyme site at the 5Ј end and an XhoI restriction enzyme site at the 3Ј end. The amplified PCR product was digested with FastDigest NheI and XhoI (Thermo Fisher Scientific), and ligated into pRSET-A (Thermo Fisher Scientific) cut with the same enzymes. The N-terminal His 6tagged proteins were then expressed in E. coli strain BL21(DE3) pLysS (Promega). Cells were cultured at 37°C in 50 ml of LB medium to an OD 600 of between 0.6 and 0.8, followed by a 4-h induction with 1 mM isopropyl-␤-D-thiogalactopyranoside at 30°C. Cells were collected by centrifugation at 5,000 ϫ g for 20 min at 4°C, frozen in liquid nitrogen, and lysed in 2 ml Bug-Buster Master Mix (EMD4 Biosciences). The tagged proteins were affinity-purified using nickel-nitrilotriacetic acid-agarose beads (Qiagen), as described previously (35), and equal amounts of protein, as determined by the Pierce BCA protein assay kit (Thermo Fisher Scientific), were subjected to SDS-PAGE and immunoblot analysis with anti-RPT4a and anti-His 6 antibodies.

Reverse transcription-PCR analysis
Total RNA was extracted from 50 to 100 mg of 10-day-old plate-grown or 7-day-old liquid-grown seedlings using the RNeasy plant mini kit (Qiagen), according to the manufacturer's instructions. Following quantification with a NanoDrop 1000 spectrophotometer (Thermo Fisher Scientific), 1 g of total RNA was treated with DNase I (Thermo Fisher Scientific), and RNA integrity was assessed by a combination of OD 260 / OD 280 and OD 260 /OD 230 measurements and by running samples on denaturing formaldehyde agarose gels. RNA was converted into cDNA using the SuperScript III first-strand synthesis system (Thermo Fisher Scientific) and oligo(dT) 20 primers, again according to the manufacturer's instructions. Following first-strand synthesis, the cDNA was diluted 1:30, and 5 l was then amplified in a 20-l reaction volume also containing 10 l of EconoTaq Plus Green Master Mix (Lucigen), 3 l of sterile H 2 O, and 1 l each of 10 mM forward and reverse primers. Details of all oligonucleotide primers are provided in Table S3.

Glycerol gradient centrifugation
Glycerol gradient fractionation of 26S proteasome sub-complexes was performed essentially as described previously (35), with minor modifications. Seedlings were grown in 5 ml of liquid GM medium at 21-23°C under continuous light for 10 days with gentle shaking (90 rpm). Typically, ϳ30 mg of dry seeds were used per culture, resulting in ϳ2 g of fresh weight tissue. Frozen tissue was ground to a powder in a mortar and pestle at liquid nitrogen temperatures, and proteins were extracted on ice for 20 min in 1 volume of Buffer A (20 mM HEPES-KOH (pH 7.5), 5 mM MgCl 2 , 10% (v/v) glycerol, 2 mM ATP, 1 mM DTT, 2 mM PMSF, 6 M chymostatin, 1ϫ plant protease inhibitor mixture, 10 mM phosphocreatine, 1 mg/ml creatine phosphokinase (all from Sigma)). Extracts were filtered through two layers of Miracloth (Calbiochem) and clarified at 30,000 ϫ g for 20 min at 4°C, and 1 ml of supernatant was loaded onto an 11-ml, 10 -40% (v/v) glycerol density gradient in Buffer A, made using an Auto-DensiFlow density gradient fractionator (Labconco). Following centrifugation at 100,000 ϫ g for 18 h at 4°C, 500-l fractions were collected manually, and 10 l of each fraction were subjected to SDS-PAGE and immunoblot analysis as described above.

Proteasome affinity purifications
Affinity purification of 26S proteasomes and their sub-complexes from Arabidopsis seedlings was performed essentially as described previously (29,64), with minor modifications. Seeds were liquid-phase-sterilized following a 30-min imbibition in sterile water, vernalized at 4°C for 4 -5 days, and then grown in 50 ml of liquid GM medium at 21-23°C under continuous light for 10 days with gentle shaking (90 rpm). Where indicated, the medium was removed on the 9th day and replenished with fresh medium containing 50 M MG132 (or an equivalent volume of DMSO as the control), with seedlings treated for an additional 16 h prior to harvesting. Typically, ϳ100 mg of dry seeds were used per culture, resulting in ϳ5 g of fresh weight tissue. Tissue was harvested, pressed dry with paper towels, and frozen in liquid nitrogen.
Frozen seedlings were ground to a powder in a mortar and pestle at liquid nitrogen temperatures as above, and proteins were extracted with 1.25 volumes of Buffer B (50 mM HEPES-KOH (pH 7.5), 50 mM NaCl, 10 mM MgCl 2 , 10% (v/v) glycerol, 5 mM DTT, 2 mM PMSF, 6 M chymostatin, 1ϫ plant protease inhibitor mixture) either containing or omitting 20 mM ATP. Extracts were filtered through two layers of Miracloth and clarified at 30,000 ϫ g for 20 min at 4°C. The supernatant was immediately applied three times over a 12-ml PolyPrep chromatography column (Bio-Rad) containing 100 l (equal to a 50-l bead volume) of anti-FLAG M2 affinity resin (Sigma). The column was washed three times with 2 ml of Buffer B (containing or omitting 20 mM ATP as required), and remaining bound protein was eluted by incubating the beads for 30 min at 4°C with 250 l of Buffer B containing 20 mM ATP and 500 ng/l of the FLAG peptide (DYKDDDDK, synthesized by the University of Wisconsin Biotechnology Center Peptide Synthesis Facility). Samples of the crude extract, flow-through (both diluted 1:10), third wash step, and/or elution were analyzed by native or SDS-PAGE (64), followed by silver staining or immunoblotting with antibodies against various proteasome subunits and ubiquitin (see above).

Tandem mass spectrometry sample preparation
For in-solution trypsin digestion, 200 l of eluant from each proteasome affinity purification was vacuum-dried to a final volume of ϳ25 l and denatured in 8 M urea, 25 mM (NH 4 )HCO 3 in a total volume of 300 l. Proteins were then reduced with 10 mM DTT at room temperature for 1 h and alkylated in the dark in the presence of 50 mM 2-chloroacetamide at room temperature for a further 1 h. Excess alkylating agent was quenched with 50 mM DTT for 5 min at room temperature, and the samples were diluted with 1.2 ml of 25 mM (NH 4 )HCO 3 to reduce the urea concentration to below 1.5 M. Proteins were digested overnight at 37°C with 1 g of sequencing grade modified porcine trypsin (Promega). Peptides were vacuum-dried as above to a final volume of ϳ300 l, acidified with 10% (v/v) trifluoroacetic acid (TFA) until the pH was less than 3.0, and desalted and concentrated on a 100-l Bond Elut Identification of plant proteasome assembly chaperones OMIX C18 pipette tip (Agilent Technologies), according to the manufacturer's instructions. Peptides were eluted in 50 l of 75% (v/v) acetonitrile, 0.1% (v/v) acetic acid, then lyophilized and resuspended in 30 l of 5% (v/v) acetonitrile, 0.1% (v/v) formic acid.
For in-gel trypsin digestion of samples after native-PAGE, silver-stained protein bands were cut into 1-mm 3 pieces, washed once with H 2 O, and destained with a freshly prepared 1:1 mixture of 100 mM sodium thiosulfate (Na 2 S 2 O 3 ) and 30 mM potassium ferricyanide (K 3 Fe(CN) 6 ) until colorless, followed by two washes with sterile distilled H 2 O to remove excess silver. Gel pieces were then dehydrated with 50% (v/v) acetonitrile in 25 mM NH 4 HCO 3 , vacuum-dried as above, and then rehydrated with reducing solution (50 mM NH 4 HCO 3 , 25 mM DTT) for 20 min at 55°C. Gel pieces were next treated with alkylating solution (50 mM NH 4 HCO 3 , 55 mM iodoacetamide) in the dark for 20 min at room temperature, washed once with sterile distilled H 2 O, and then dehydrated and vacuum-dried as above. Proteins were digested overnight at 37°C with 1 g of sequencing grade modified porcine trypsin (Promega) together with ProteaseMAX surfactant (66). Peptides were vacuum-dried and acidified as above, desalted with self-packed C18 stage-tips using a spin protocol (67), and then lyophilized and resuspended in 30 l of 5% (v/v) acetonitrile, 0.1% (v/v) formic acid.

Tandem mass spectrometry instrument parameters
Nano-scale ultra-high-performance liquid chromatographic (UHPLC) separation of tryptic peptides digested in solution was performed on a Dionex Ultimate 3000 Rapid Separation LC system (Thermo Fisher Scientific). The protein digests were loaded onto a 1-l nano-Viper sample loop (Thermo Fisher Scientific) and separated on an analytical C18 column (Acclaim PepMap RSLC C18 column with a 2-m particle size, 100-Å pore size, 75 m ϫ 15 cm (Thermo Fisher Scientific)) by the application of a 90-min linear gradient from 2 to 32% (v/v) acetonitrile in 0.1% (v/v) formic acid, with the column flow rate set to 300 nl/min. MS analysis of the eluted tryptic peptides was performed online using a Q Exactive Plus mass spectrometer (Thermo Fisher Scientific) possessing a Nanospray Flex ion source (Thermo Fisher Scientific) fitted with a stainless steel nano-bore emitter operated in positive electrospray ionization (ESI) mode at a capillary voltage of 1.6 kV. Data-dependent acquisition of full MS scans within a mass range of 380 -1,500 m/z at a resolution of 70,000 was performed, with the automatic gain control (AGC) target set to 1 ϫ 10 6 ion intensity and the maximum fill time set to 100 ms. High-energy collision-induced dissociation (HCD) fragmentation of the top 15 most intense peaks was performed with a normalized collision energy of 28, an intensity threshold of 1 ϫ 10 5 counts, and an isolation window of 1.2 m/z, excluding precursors that had unassigned, ϩ1, ϩ7, or ϩ8 charge states. MS/MS scans were acquired with a mass range of 200 -2,000 m/z at a resolution of 17,500, with an AGC target of 8 ϫ 10 3 and a maximum fill time of 80 ms. All peaks were recorded in profile mode. Dynamic exclusion was enabled with a repeat count of 2 and an exclusion duration of 10 s, and the minimum MS ion count for triggering MS/MS was set to 8 ϫ 10 3 counts.
Peptide samples obtained from in-gel digestion were instead analyzed using a nanoAcquity UHPLC system (Waters) connected online to an LTQ Velos Orbitrap mass spectrometer (Thermo Fisher Scientific) operated in positive ESI mode. The protein digests were separated on a fused silica capillary C18 micro-column (Waters; 1.7-m particle size, 130-Å pore size, 100 m ϫ 20 cm, with an emitter tip pulled to ϳ1 m) by the application of a linear gradient from 2 to 30% (v/v) acetonitrile in 0.1% (v/v) formic acid, with the column flow rate set to 300 nl/min. Data-dependent acquisition of full MS scans within a mass range of 300 -1,500 m/z at a resolution of 60,000 was performed, followed by HCD fragmentation of the top 10 most intense peaks, performed with a normalized collision energy of 42, an intensity threshold of 3 ϫ 10 3 counts, and an isolation window of 3 m/z, excluding precursors that had unassigned or ϩ1 charge states. MS/MS scans were acquired with a mass starting at 100 m/z at a resolution of 7,500, with an AGC target of 1 ϫ 10 5 and a maximum fill time of 500 ms. Dynamic exclusion was enabled with a repeat count of 2, an exclusion duration of 30 s, and an exclusion window of 2 min, whereas the minimum MS ion count for triggering MS/MS was set to 3 ϫ 10 3 counts. Data were acquired in centroided mode.

Tandem mass spectrometry data processing
Raw MS/MS spectral files were processed with Max-Quant version 1.5.3.30 (68), set up to use the Andromeda search engine to interrogate the Arabidopsis proteome file TAIR10_pep_20101214_updated.fasta, obtained from The Arabidopsis Information Resource (TAIR) database version 10.1, which contained 35,518 entries (Table S4). Search parameters were set to assume trypsin digestion with a maximum of two missed cleavages, precursor mass tolerances of 20 ppm, and fragment mass tolerances of 10 ppm. Carbamidomethylation of cysteine was specified as a static modification, whereas oxidation of methionine, N-terminal acetylation, and Gly-Gly footprints (i.e. the ubiquitylation remnant on a lysine residue after cleavage with trypsin) were specified as dynamic modifications. The target FDR of 0.01 was used as validation for peptide-spectral matches and peptides. Ten proteins were marked as likely contaminants, including nitrilases, cruciferins, and seed storage albumins (Table S5).
LFQ analyses were performed using the MaxLFQ algorithm in MaxQuant (36) with default settings, except that only unique peptides were used for quantification so that protein isoforms could be discriminated. Matching was set to "match from and to samples" to increase the number of quantifiable proteins per run, using the "match between runs" feature in MaxQuant. The resulting proteingroups.txt files (Table S6) were processed using Perseus (69). MaxLFQ values were averaged for both technical replicates, and missing MaxLFQ quantification values were imputed with values representing a normal distribution around the detection limit of the mass spectrometer, i.e. the mean and standard deviation for the distribution of the real intensities were determined, then a new distribution with a downshift of 1.8 standard deviations and a width of 0.3 standard deviations was created. The total matrix was then imputed using these values, enabling statistical analysis. Volcano plots comparing samples were generated in Perseus, with parameters Identification of plant proteasome assembly chaperones set to use a two-sided t test, an FDR of 0.01, an S0 value of 2, and 250 permutations, and then plotted using the Seaborn Python graphing library (http://doi.org/10.5281/zenodo. 824567). KDEs were also calculated in Seaborn, whereas the median and Kolmogorov-Smirnov (KS) tests for enrichment were performed using SciPy Statistical Functions version 1.0. For the bar graphs in Fig. 3, missing values were imputed with a log 2 (MaxLFQ value) of 15 so that a "zero value" could be more easily viewed along the y axis.
For analysis of native-PAGE gel slices, MS/MS spectra were analyzed using the Morpheus search engine (70), set up to search the same proteome database as above. Search parameters assumed digestion with trypsin, a maximum of two missed cleavages, a minimum peptide length of 6, precursor mass tolerances of 2.10 Da, and fragment mass tolerances of 0.01 Da. Static and dynamic modifications were specified as above, and the FDR of 0.01 was used as validation for peptide-spectral matches and peptides. LFQ was performed with Morpheus Spectral Counter (51) based on dNSAF values (50 ; Tables S7  and S8). The data were plotted in R using the heatmap.2 function, with centroided hierarchical clustering based on Pearson's correlation used as the distance function.

Gene expression analysis
Raw transcript abundance data (excluding UMP1b, for which no data were available) were downloaded from the Transcriptome Variation Analysis database (www.travadb.org) 7 (47) and plotted in R using the heatmap.2 function as above. All read counts were normalized to values ranging from 0 to 1 by the "median-of-ratios" method as in DESeq2 (71), and then divided by the maximum expression for each gene. Data for gene expression in response to MG132 treatment or in the rpn10 -1 and rpn12a-1 mutants were obtained from a previously published RNA-seq dataset (GSE81668 (48)), and the heat map was plotted with Perseus (69).

Yeast complementation assays
Unless otherwise noted, all yeast manipulations were performed according to standard protocols. Details of all strains used in this study are given in Table S9, with strains BY4741 or BY4742 used as the WT controls (72). Deletion strains (⌬pba3, ⌬pba4, and ⌬ump1) were obtained from the yeast knockout collection (Dharmacon) and cultured on YPDA medium containing 200 mg/ml geneticin. All genomic deletions were confirmed by PCR genotyping, using primer pairs A ϩ B, A ϩ KanB, C ϩ D, KanC ϩ D, and A ϩ D for each deletion collection strain, sequences of which are provided in Table S3.
To clone required coding sequences, total RNA was first extracted from Arabidopsis seedlings or yeast cells and then converted into cDNA. Arabidopsis RNA was extracted from 50 to 100 mg of 7-day-old liquid-grown seedlings as described above. To obtain total yeast RNA, the cell walls of ϳ2 ϫ 10 7 freshly harvested BY4742 cells (taken from a 15-ml culture grown in YPDA) were digested with 100 units of lyticase (from Arthrobacter luteus) in 100 ml of buffer Y1 (1 M sorbitol, 100 mM EDTA, 0.1% (v/v) ␤-mercaptoethanol (pH 7.4)) for 1 h at 30°C, and RNA was then extracted using the RNeasy mini kit (Qiagen). Total RNA was then converted into first-strand cDNA as described above. Coding sequences amplified by PCR were first recombined into pDONR221 via the Gateway BP clonase II reaction (Thermo Fisher Scientific), with the sequence for an N-terminal FLAG or HA tag incorporated into the appropriate PCR amplification primers. Sequence-confirmed genes were then recombined into the pAG423-GPD-ccdB, pAG425-GPD-ccdB, or pAG426-GPD-ccdB vectors (Addgene, product numbers 14150, 14154, and 14156, respectively) via the Gateway LR clonase II reaction (Thermo Fisher Scientific), and the resulting plasmids (or the empty vectors as controls) were transformed into the indicated yeast strains using the standard lithium acetate procedure. Transformed cells were cultured on synthetic dropout medium lacking histidine and/or leucine as appropriate.
For growth assays, cells were grown in 10 ml of synthetic dropout medium lacking histidine, leucine, and/or uracil. Following a 12-h overnight growth, cultures were diluted to an OD 600 of 1.0 in the same medium then subjected to a series of 5-fold dilutions, before 5 l of each dilution was spotted onto the same medium containing or lacking 5 M canavanine. Cells were then grown for 36 h at 30 or 37°C prior to imaging.

Sequence alignments and phylogenetic analyses
The predicted full-length nucleotide and protein sequences of A. thaliana PBAC1, PBAC2, PBAC3, PBAC4, UMP1a, HSM3, NAS2, NAS6, and ECM29 (obtained from the Arabidopsis Information Resource database version 10.1) were used as queries in BLAST searches for related loci in other plant genomes available in the Joint Genome Initiative's Phytozome database (www.phytozome.net). 7 Related animal and yeast sequences were identified as the top hits in each respective proteome resulting from PSI-BLAST searches using the BLO-SUM62 substitution matrix, provided that the default expectation value (E-value) was above the threshold of 0.005 (73). Domain positions were predicted by the SMART (www. smart.embl-heidelberg.de) 7 and PFAM (www.pfam.sanger.ac. uk) 7 databases.
Progressive alignments of the predicted full-length amino acid sequences were performed using Clustal Omega (www. clustal.org/omega) 7 with the default settings. Following minor manual editing, the final alignments were displayed with Box-Shade version 3

Identification of plant proteasome assembly chaperones
␥-distributed rate variation with a proportion of invariable sites. The software was run for 1,000,000 generations, with a sampling frequency of every 1,000 generations. The first 250,000 generations (25%) were discarded as "burn-in" after checking that the log likelihood values had plateaued and that the potential scale reduction factor was close to 1.000. The resulting consensus trees were displayed using FigTree version 1.4.2 (www.tree.bio.ed.ac.uk/software/figtree). 7

Statistical analyses
Details of MS-based statistical analyses are provided above. All other datasets were statistically analyzed using one-way analysis of variance (ANOVA) followed by Tukey's post-hoc tests to identify significantly different data points. At least three biological replicates were performed in all cases, unless otherwise indicated in the figure legends.

Accession numbers
All accession numbers for genes and proteins used in this study are given in Table S10. The MS-based proteomics data have been deposited to the ProteomeXchange Consortium (www.proteomexchange.org) 7 via the PRIDE partner repository with the dataset identifier PXD014610. The sample to which each .raw file corresponds is given in Table S11.