The eta7/csn3-3 Auxin Response Mutant of Arabidopsis Defines a Novel Function for the CSN3 Subunit of the COP9 Signalosome

The COP9 signalosome (CSN) is an eight subunit protein complex conserved in all higher eukaryotes. In Arabidopsis thaliana, the CSN regulates auxin response by removing the ubiquitin-like protein NEDD8/RUB1 from the CUL1 subunit of the SCFTIR1/AFB ubiquitin-ligase (deneddylation). Previously described null mutations in any CSN subunit result in the pleiotropic cop/det/fus phenotype and cause seedling lethality, hampering the study of CSN functions in plant development. In a genetic screen to identify enhancers of the auxin response defects conferred by the tir1-1 mutation, we identified a viable csn mutant of subunit 3 (CSN3), designated eta7/csn3-3. In addition to enhancing tir1-1 mutant phenotypes, the csn3-3 mutation alone confers several phenotypes indicative of impaired auxin signaling including auxin resistant root growth and diminished auxin responsive gene expression. Unexpectedly however, csn3-3 plants are not defective in either the CSN-mediated deneddylation of CUL1 or in SCFTIR1-mediated degradation of Aux/IAA proteins. These findings suggest that csn3-3 is an atypical csn mutant that defines a novel CSN or CSN3-specific function. Consistent with this possibility, we observe dramatic differences in double mutant interactions between csn3-3 and other auxin signaling mutants compared to another weak csn mutant, csn1-10. Lastly, unlike other csn mutants, assembly of the CSN holocomplex is unaffected in csn3-3 plants. However, we detected a small CSN3-containing protein complex that is altered in csn3-3 plants. We hypothesize that in addition to its role in the CSN as a cullin deneddylase, CSN3 functions in a distinct protein complex that is required for proper auxin signaling.


Introduction
The phytohormone auxin (indole-3-acetic acid or IAA) regulates numerous plant developmental processes, by modulating the expression of auxin responsive genes to control cell division, expansion and differentiation [1]. Two large families of transcriptional regulators play essential roles in auxin mediated gene expression: short-lived Aux/IAA proteins and Auxin Response Factor (ARF) transcription factors. In the absence of an auxin signal, Aux/IAAs heterodimerize with ARFs to repress the transcriptional activation of auxin responsive genes [2,3]. To activate auxin responsive gene expression, Aux/IAA repression must be removed, which is accomplished by the SCF TIR1/AFB E3 ubiquitin ligase and the 26S proteasome mediated proteolysis of Aux/IAAs in an auxin-dependent manner [4][5][6].
Ubiquitin is a small conserved protein, which can be covalently conjugated to specific substrate proteins through the activity of a ubiquitin-activating enzyme (E1), a ubiquitin-conjugating enzyme (E2), and a ubiquitin-ligase (E3) [7]. Proteins with a polyubiquitin chain can then be recognized and degraded by the 26S proteasome. A major class of E3 ubiquitin ligases are the cullin-RING ligases (CRLs), among which the SCF (for Skp1-Cullin-F-box) sub-type has been well-studied [8]. SCF ligases consist of four subunits: the cullin protein, CUL1, binds the RING box protein RBX1 with its C-terminus, while the SKP1 adaptor protein (ASK1 in Arabidopsis) recruits any of many F-box proteins (FBPs) to the N-terminus of CUL1 [9,10]. The TIR1, AFB1, AFB2, and AFB3 F-box proteins of the SCF TIR1/AFB complex are auxin receptors, targeting Aux/IAAs for ubiquitination and degradation upon binding auxin [11][12][13]. Mutations in SCF TIR1/AFB subunits, such as axr6/cul1, ask1 and tir1, cause the stabilization of Aux/ IAAs and a variety of auxin response defects including auxin resistant root growth and reduced lateral root development [4,14,15]. The proper function of SCF ligases, as well as other CRLs, also requires a process called neddylation, in which the ubiquitin-like protein RUB/NEDD8 is covalently conjugated to the cullin subunit [16][17][18]. The AXR1, ECR1, and RCE1 enzymes mediate cullin neddylation, and mutations in these factors result in a reduction in the level of neddylated cullin and pleiotropic growth defects, including diminished auxin signaling [16,19]. The NEDD8 modification of CUL1 is a highly dynamic process [20,21]. Cleavage of NEDD8 from cullins (deneddylation) is promoted by a protein complex called the COP9 signalosome (CSN), as the CSN complex purified from porcine spleen deneddylates Cul1 in vitro [21,22]. Consistently, studies in various organisms have shown that neddylated cullin proteins accumulate in csn mutants [22][23][24][25].
The COP9 signalosome (CSN) is an evolutionarily conserved protein complex of eight subunits [26]. It was originally identified in Arabidopsis through genetic screens for mutants exhibiting a constitutive photomorphogenic/de-etiolated (cop/det) phenotype, and was subsequently biochemically purified from both plant and animal protein extracts [26,27]. The CSN is structurally related to the 19S lid of the 26S proteasome and eukaryotic translation initiation factor 3 (eIF3), and is composed of six PCI (for Proteasome, COP9, eIF3) domain-containing subunits (CSN1-4, 7, and 8) and two MPN (for Mpr1p, Pad1p N-terminal) domaincontaining subunits (CSN5 and CSN6) [27]. It has been established that the CSN regulates SCF activity by deneddylating the CUL1 subunit of the SCF [22,28]. Besides CUL1, the CSN also deneddylates other cullin proteins and broadly regulates many CRLs, including those containing CUL2, CUL3 and CUL4 [21,23,29,30]. In plants, the CSN has been implicated in a variety of processes, including auxin, jasmonate (JA), and gibberellin (GA) signaling, flower development, and light signaling via its interaction with many CRLs [29,31,32]. CSN-mediated cullin deneddylation is catalyzed by the JAMM metallozyme motif within the CSN5 subunit [22,33]. However, CSN5 can only provide this isopeptidase activity after incorporation into the CSN holocomplex, as deneddylation activity is defective in all previously characterized CSN subunit mutants [23,34,35].
Neddylation plays a positive role in regulating SCF activity, by promoting a conformational change in CUL1 that shortens the distance between FBP-bound substrates and the E2 ubiquitinconjugating enzyme [36][37][38]. Although in vitro biochemical studies indicated that the CSN negatively regulates E3 ubiquitin ligase activity [21,33], genetic evidence suggests otherwise [28,39,40]. Studies in yeast and Drosophila suggested that although the CSN returns CRL activities to basal levels, it can also facilitate CRL activities by either maintaining the stability of labile substrate adapters or by recycling unstable, neddylated cullins into the more stable deneddylated isoform [41,42]. In Arabidopsis, the CSN is also required for the proper functions of SCF TIR1/AFB , as well as SCF SLY1 , in the degradation of Aux/IAA proteins or the DELLA proteins, respectively [28,31,43]. Together, these findings suggest that cullin neddylation/deneddylation is a highly dynamic process essential for maintaining proper CRL function [44,45]. In addition to regulating SCF activity by deneddylating cullins, recent biochemical findings indicate that the CSN also inhibits SCF complexes by a noncatalytic mechanism [46,47]. Following deneddylation, the CSN remains stably associated with the SCF, sterically hindering both the re-neddylation of CUL1 and SCF interactions with the E2 enzyme.
Arabidopsis null mutants of any CSN subunit are phenotypically indistinguishable and exhibit the pleiotropic cop/det/fus phenotype, which is characterized by short hypocotyl and open cotyledons of dark-grown seedlings, accumulation of anthocyanin and seedling lethality [23,27,48,49]. Severe deneddylation defects have been found in previously described null mutants of each of the CSN subunits [23,48], but the associated seedling lethality hampers the further analyses of CSN functions. Recently however, a few weak, viable Arabidopsis csn mutants have been described. These include mutants lacking one of the two copies of CSN5 and CSN6 encoded in the Arabidopsis genome [23,50], as well as hypomorphic missense alleles of CSN1 and CSN2 [40,43]. Importantly however, these viable csn mutants still exhibit diminished deneddylation activity, resulting in the accumulation of neddylated CUL1 and reduced SCF activity [23,28,34,35,40,43].
In a previously described genetic screen for enhancers of tir1-1 [15,[51][52][53][54], we identified two weak csn subunit mutants, designated as eta6/csn1-10 [43] and eta7/csn3-3. Our phenotypic characterization of these csn mutants, together with expression studies with auxin regulated reporters, demonstrate that csn1-10 and csn3-3 confer very similar reductions in auxin response. However, unlike csn1-10, which is a typical csn mutant with defects in CSNmediated deneddylation and Aux/IAA protein degradation [43], neither of these defects were observed in the csn3-3 mutant. Furthermore, genetic interactions between these csn mutants and additional auxin signaling mutants also distinguish csn3-3 from other csn mutants. Our studies suggest that csn3-3 is a unique csn mutant that defines a novel functional activity for the CSN3 subunit of the COP9 signalosome in the regulation of auxin signaling.

Results
eta7 is a weak allele of COP9 signalosome subunit 3 (CSN3) We have previously reported the identification of several auxin response mutants isolated from a genetic screen for enhancers of the tir1-1 auxin response defect (designated as eta mutants), including eta1/axr6-3 [15], eta2-1/cand1 [52], eta3/sgt1b [53], eta4/ pdr9-1 [54], eta5/iar4 [51] and eta6/csn1-10 [43]. The eta7 mutant was also identified in this screen. Map based cloning positioned the eta7 mutation within a 330 kb interval on chromosome five. This interval contained 102 predicted genes, including CSN3/FUS11, which encodes subunit 3 of the COP9 signalosome (CSN) [55]. Given that csn subunit mutations are known to confer diminished auxin response phenotypes, we sequenced the CSN3 locus from eta7 plants and identified a single mutation ( Figure 1A). This mutation results in a G293E missense mutation within the highly conserved PCI domain of CSN3. This domain is important for subunit interaction and CSN complex assembly [56]. Primary sequence alignment among the PCI domains of several CSN3 orthologs revealed that Gly293 is extremely highly conserved throughout eukaryotes ( Figure 1B). To confirm that this mutation was responsible for the eta7 auxin response defect, we conducted complementation tests by transforming a genomic CSN3 construct into eta7 mutant plants. The CSN3 transgene fully restored auxin sensitivity to eta7 seedlings when tested in root growth assays ( Figure 1C). We therefore renamed eta7 as csn3-3.

Characterization of csn3-3 phenotypes
The CSN plays a well-established role in auxin signaling, acting as a deneddylase to regulate SCF TIR1/AFB activity. Mutants in any CSN subunit exhibit auxin related phenotypes, such as auxin resistant root growth and reduced lateral root formation [28,34,43]. In a dose-response assay measuring the auxin inhibition of root elongation, we found csn3-3 was mildly resistant to exogenous auxins. After transfer to media supplemented with 0.05 mM 2,4-D, root growth of wild-type (Col) seedlings was nearly completely inhibited ( Figure 2A). However, csn3-3 seedlings were resistant to this inhibition, displaying only 47% inhibition of root growth. A similar degree of auxin resistant root growth was observed with the csn1-10 and tir1-1 mutants. Additionally, both csn1-10 and csn3-3 enhanced the weak auxin resistance phenotype of tir1-1, with csn1-10 tir1-1 and csn3-3 tir1-1 double mutants exhibiting comparable dose-response profiles in root growth inhibition assays (Figure 2A). Similar assays using IAA-supplemented media also found that csn3-3 exhibited auxin resistance and enhanced the tir1-1 phenotype, demonstrating that the auxin response defect of csn3-3 is not 2,4-D specific ( Figure 2B).
Previous reports have shown that plants expressing a CSN5 antisense construct develop fewer lateral roots and reduced root hair elongation compared to wild-type controls [28]. Consistent with this finding, the csn3-3 and the csn1-10 mutations enhanced the weak lateral root defect of tir1-1 seedlings, with both double mutants developing fewer than 50% of the number of lateral roots observed in tir1-1 seedlings ( Figure 2C). Furthermore, while lateral root numbers in csn3-3 and csn1-10 single mutants were comparable to wild-type controls, csn3-3 csn1-10 double mutants developed significantly fewer lateral roots, suggesting that auxin response is more severely impaired in the double mutant background.
To further compare the auxin response defects of csn3-3 and csn1-10 mutants, we introduced the BA3:GUS and DR5:GUS auxin responsive reporters [3,57] to examine auxin mediated gene expression. Treatment of Col [BA3:GUS] control seedlings with 2,4-D or IAA triggered a strong GUS signal in the root elongation zone. In contrast, only a slight induction of BA3:GUS activity was observed in csn3-3 and csn1-10 mutants ( Figures 2D and S1). Like the auxin-resistant root growth phenotype of csn3-3, the diminished BA3:GUS expression was complemented by a genomic CSN3 transgene ( Figure 2E). Similar findings were obtained with csn3-3 and csn1-10 seedlings carrying the DR5:GUS reporter ( Figure 2D). Together, our analysis of root growth inhibition, lateral root development, and auxin mediated gene expression suggest that the csn3-3 and csn1-10 mutations confer a very similar reduction in auxin sensitivity.
CUL1 deneddylation and Aux/IAA stability are not affected in csn3-3 seedlings The CSN cleaves the RUB/NEDD8 peptide from the cullin subunit of CRL ubiquitin-ligases. All previously described csn mutants, including the weak csn1-10 and csn2-5 alleles, result in an increase in the CUL1-NEDD8 to unmodified CUL1 ratio [40,43]. We therefore examined if the csn3-3 mutation affected CUL1 modification. We included csn1-10 for comparison with csn3-3, since both mutants are viable csn mutants and display similar auxin response defects ( Figure 2). While a clear increase in CUL1-NEDD8:CUL1 was seen in extracts prepared from csn1-10 seedlings, no change from wild-type was detected in a-CUL1 western blots of csn3-3 extracts ( Figure 4A). In contrast, CUL1 was exclusively in its neddylated form (CUL1-NEDD8) in csn3-2 extracts ( Figure 4A), as reported in previous studies [35].
Since the csn3-3 mutation unexpectedly did not affect the CUL1-NEDD8:CUL1 ratio, we examined double mutants with csn1-10 to test the possibility that csn3-3 might enhance the weak deneddylation phenotype of csn1-10. Once again however, the csn3-3 mutation did not increase the CUL1-NEDD8:CUL1 ratio ( Figure 4A). Since the CSN also regulates other cullin-based E3 ubiquitin-ligases by deneddylating their cullin subunit, we applied a similar immunoblotting assay to examine CUL4 modification in csn3-3 seedlings. Similar to our findings with CUL1, we observed no accumulation of the CUL4-NEDD8 isoform in csn3-3 extracts, whereas neddylated CUL4 was readily detectable in csn1-10 and csn5a-2 extracts ( Figure 4B). Together, our analyses of CUL1 and CUL4 suggest that the csn3-3 mutation does not affect the deneddylation activity of the CSN.
The reduced deneddylation activity of several previously characterized csn mutants has been found to result in diminished SCF TIR1/AFB activity, resulting in the stabilization of Aux/IAA proteins [28,40]. Since our finding that csn3-3 plants exhibited no change in cullin deneddylation was quite surprising, we examined SCF TIR1/AFB activity by monitoring Aux/IAA stability using the previously described HS:AXR3NT-GUS reporter protein [4]. csn1-10 seedlings were again included for comparison. 6-d.o. Col, csn1-10 and csn3-3 seedlings, all of which carried the HS:AXR3NT- GUS construct, were heat-shocked at 37uC for 2 hours to induce expression, followed by return to ambient temperature and treatment with auxin. AXR3NT-GUS activity was examined both qualitatively and quantitatively at 20 min intervals during the treatment to monitor the remaining levels of AXR3NT-GUS fusion protein ( Figure 4C, D). Following the 2h heat-shock induction, similar AXR3NT-GUS levels were observed in wildtype, csn3-3 and csn1-10 seedlings ( Figure 4C). During the auxin treatment, AXR3NT-GUS activity in csn3-3 and wild-type seedlings diminished with comparable kinetics ( Figure 4D), suggesting that SCF TIR1/AFB activity is unaffected in csn3-3 plants.
To further confirm that csn3-3 had no effect on SCF TIR1/AFBmediated proteolysis, we examined another Aux/IAA reporter, IAA28-myc [60]. The IAA28-myc construct was crossed into csn3-3 and csn1-10 backgrounds. Abundance of the IAA28-myc protein was then examined by treating seedlings with IAA and immunoblotting root protein extracts. As previously reported [60], IAA28myc was rapidly degraded in wild-type seedlings and was nearly undetectable after a 10 minute auxin treatment ( Figure S2). While IAA28-myc was clearly more stable in csn1-10 seedlings, it was rapidly degraded in the csn3-3 background ( Figure S2). Combined with our HS:AXR3NT-GUS degradation data, this strongly suggests that SCF TIR1/AFB activity is unaltered by the csn3-3 mutation.
Genetic interactions distinguish csn3-3 and csn1-10 The CSN regulates auxin signaling by deneddylating CUL1 to modulate SCF TIR1/AFB activity [22,28,44]. Our findings with the csn1-10 mutant are consistent with this model. The csn3-3 mutation on the other hand, affects neither CUL1 deneddylation nor SCF TIR1/AFB -mediated Aux/IAA degradation, yet confers auxin response defects virtually identical in severity to csn1-10. These findings strongly suggest that the CSN, or at least the CSN3  subunit, plays a second role in the regulation of auxin signaling in addition to deneddylating CUL1. If so, we reasoned that csn3-3 and csn1-10 may exhibit distinct genetic interactions when combined with other mutations affecting auxin signaling.
In contrast to the double mutants with axr1-12 and axr6-3, the phenotypic severity was reversed when csn1-10 and csn3-3 were combined with the eta2-1 mutation. eta2-1 is a missense mutation in CAND1, a cullin binding protein that mediates cycles of SCF complex assembly and disassembly [43,52]. While the eta2-1 mutation abolishes CUL1 binding activity and thus disrupts SCF cycling, it has no detectable effect on CUL1 neddylation. Whereas csn1-10 eta2-1 double mutants failed to progress past the seedling stage [43], csn3-3 only slightly enhanced the developmental defects of eta2-1 ( Figure 5F). Together, the highly differential genetic interactions conferred by the csn1-10 and csn3-3 mutations in combination with axr1-12, axr6-3, and eta2-1 strongly suggest that these two mutations in CSN subunits affect distinct aspects of auxin signaling. This notion is consistent with our finding that only csn1-10 exhibits defects in cullin deneddylation and SCF TIR1/AFBmediated regulation of Aux/IAA protein stability.
A novel CSN3 complex, but not the CSN holocomplex, is altered in csn3-3 plants The CSN deneddylase activity is catalyzed by the CSN5 subunit, in which the JAMM motif is the catalytic center required for the isopeptidase activity to remove NEDD8 from the cullin proteins [33,44]. However, the proper assembly of the CSN holocomplex is required for deneddylase activity. Consequently, null mutations in any of the eight CSN subunits abolish both CSN assembly and deneddylase activity, and confer similar seedlinglethal phenotypes [23,[48][49][50]55,59].
Since csn3-3 confers reduced auxin signaling but does not affect the deneddylase activity of the CSN, we examined whether the csn3-3 mutation had any effect on CSN complex assembly. Protein extracts from Col and csn3-3 seedlings were fractionated by gel filtration chromatography as previously described [43], and analyzed by immunoblotting with antibodies against several CSN subunits. With wild-type extracts, the CSN1, CSN4, and CSN8 subunits were detected primarily in high molecular mass fractions (#6-9) corresponding to 450-700 kD ( Figure 6A), consistent with the previously observed molecular mass of the CSN holocomplex [23,58]. No obvious differences from this pattern were detected with csn3-3 extracts. Unlike other CSN subunits, prior studies in both plants and animals have demonstrated that the CSN5 subunit is present in both the CSN and as a monomer [34,59,63]. Consistent with these findings, we detected CSN5 in both high and low molecular mass fractions corresponding the CSN holocomplex and CSN5 monomer, respectively ( Figure 6B). Again however, no differences between Col and csn3-3 were observed. In contrast, the csn1-10 mutation resulted in a clear shift in CSN5 fractionation, with the majority of CSN5 fractionating in the monomeric form ( Figure 6B). This finding suggests that the csn1-10 mutation impairs incorporation of CSN5 into the CSN holocomplex and is consistent with the reduction in cullin deneddylation we observe in this mutant.
Together with our finding that CSN deneddylase activity is unaltered in csn3-3 mutants, the above gel filtration findings with CSN1, CSN4, CSN5, and CSN8 indicate that CSN holocomplex assembly is unaffected by csn3-3. Consistent with this conclusion, we found that the abundance of the csn3-3 mutant protein is similar to wild-type, and the mutant protein also incorporates into the CSN holocomplex ( Figures 6C, D). However, during our gel filtration studies with CSN3, we noticed an additional CNS3 elution peak centered around 130 kD in wild-type extracts ( Figure 6C, fractions 12-14). Interestingly, this small CSN3containing complex (sCSN3c) was absent in csn3-3 seedlings, indicating that the csn3-3 missense mutation specifically disrupts the formation of this complex. Furthermore, the sCSN3c was restored by expression of a P CSN3 :CSN3 transgene in csn3-3 mutant plants (L4 in Figure 6C).
Our results suggest that sCSN3c represents a novel CSN3containing complex and not an intermediate in CSN holocomplex assembly. First, a prior CSN subunit interaction mapping study concluded that the CSN3 subunit primarily contacts CSN1, CSN8, and probably CSN4 [64]. However, none of these subunits co-eluted with sCNS3c in our gel filtration experiments ( Figure 6A). Likewise, no overlap in the elution profile of CSN5 with sCSN3c was observed ( Figure 6A, B). Secondly, we tested the possibility that sCSN3c may be due to partial disassembly of the CSN that occurred during our in vitro fractionations. Fractions containing the CSN holocomplex were isolated and subjected to a second round of gel filtration and subsequently immunoblotted with a-CSN3 antibody ( Figure S3). However, no CSN3 protein was detected in the low molecular mass fractions corresponding to sCSN3c, demonstrating that the CSN is stable under our conditions. Together, these findings suggest that CSN3 is the only canonical CSN subunit present in sCSN3c, although the possibility that subunits for which we do not have antibodies (CSN2/6/7) are sCSN3c components cannot be eliminated.

Discussion
In plants, animals, and fungi, the CSN has a well-established role as a cullin deneddylase that regulates CRL ubiquitin-ligase activity [44]. Prior reports in Arabidopsis have demonstrated that reduced deneddylase activity in various csn mutants affects SCF TIR1/AFB ubiquitin-ligase activity and consequently results in impaired auxin signaling [28,35,40]. In this study, we identified the csn3-3 missense mutation as an enhancer of the auxin resistant root growth phenotype conferred by tir1-1. Consistent with diminished auxin signaling, csn3-3 also enhanced the tir1-1 lateral root development defect and conferred diminished auxin mediated expression of the DR5: and BA3:GUS reporters. While these findings were hardly surprising given the aforementioned studies of Arabidopsis CSN subunit mutants, we quite unexpectedly found that neither the deneddylation activity of the CSN nor the SCF TIR1/AFB -mediated regulation of Aux/IAA protein stability was affected by the csn3-3 mutation. These findings clearly distinguish csn3-3 from all previously characterized csn mutants.
It could be argued that csn3-3 is simply a weak mutation that does not confer sufficient defects to be detected in assays examining neddylation status of cullins or the SCF TIR1/AFB activity. However, our comparison of csn3-3 phenotypes to those of a second weak CSN subunit mutant, csn1-10, strongly argues against this possibility. csn3-3 and csn1-10 confer virtually identical dose-response curves in assays examining auxin-mediated inhibition of root growth, with both mutants exhibiting 50% growth inhibition at ,50 nM 2,4-D. Likewise, both mutants enhance the tir1-1 auxin resistant root growth and reduced lateral root development phenotypes to a similar extent. Furthermore, both csn3-3 and csn1-10 confer comparable reductions in auxin mediated expression of the BA3: and DR5:GUS reporters. Together, these findings indicate that csn3-3 and csn1-10 impair auxin signaling to a similar extent. However, whereas cullin deneddylation and Aux/IAA degradation were unaffected in csn3-3 seedlings, both of these molecular defects were clearly apparent in csn1-10 mutants. Prior studies have clearly demonstrated that the CSN3 subunit is required for CSN deneddylase activity [35], which we have confirmed with the csn3-2 null allele. The csn3-3 missense mutation however, confers a reduction in auxin response without affecting cullin deneddylation, suggesting that CSN3 plays a second role in auxin signaling in addition to its role in regulating the SCF TIR1/AFB ubiquitin ligase (Figure 7).
The dramatically different double mutant phenotypes exhibited when the csn1-10 and csn3-3 mutations were combined with axr6-3, axr1-12, or eta2-1 also indicate that csn1-10 and csn3-3 affect distinct aspects of auxin signaling. While csn3-3 confers seedling lethality when combined with axr6-3 or axr1-12, the csn1-10 mutation does not. In contrast, csn1-10, but not csn3-3, confers a seedling arrest phenotype in the eta2-1 background. Given that SCF TIR1/AFB activity appears unaffected in csn3-3 plants, it seems unlikely that the lethality of csn3-3 axr1-12 and csn3-3 axr6-3 seedlings is due to a further reduction in SCF TIR1/AFB activity. Instead, we speculate that csn3-3 specifically affects auxin signaling downstream or perhaps independently of SCF TIR1/AFB (Figure 7), such that the combination of csn3-3 with axr6-3 or axr1-12 causes auxin sensitivity to fall below the threshold required for early seedling development. Although it is unclear what these differential genetic interactions might mean mechanistically, it is interesting to note that both axr6-3 and axr1-12 result in a reduction in neddylated CUL1 [15,16,19,37]. On the other hand, the eta2-1 mutation has no effect on CUL1 neddylation status [52]. Rather, the eta2-1 mutation abolishes the CUL1 binding activity of CAND1, resulting in the disruption of CAND1-mediated cycling of SCF complexes [43]. Consistent with the hypothesis that an SCF TIR1/AFB -independent pathway may regulate auxin signaling to control gene expression, like csn3-3, the previously described ibr5 mutants of Arabidopsis also exhibit diminished auxin response without inhibiting SCF TIR1/AFB -mediated degradation of Aux/IAA proteins [60]. IBR5 encodes a putative dual-specificity protein phosphatase. However, although Aux/IAA proteins are highly unstable in both csn3-3 and ibr5 mutants, ibr5 seedlings exhibited reduced steady-state levels of the AXR3NT-GUS and IAA28-myc reporter proteins. In our analysis of csn3-3 mutants, however, both of these reporter proteins were present at levels comparable to wild-type controls. Furthermore, unlike csn3-3, ibr5 does not interact with axr1 in a synergistic manner [60]. Together, these findings suggest that it is unlikely that csn3-3 and ibr5 share a common auxin signaling defect.
The fact that the csn3-3 mutation did not diminish CSN deneddylase activity, yet conferred reduced auxin response phenotypes, suggests that this mutation defines a novel function for CSN3. Therefore, a major question is whether csn3-3 defines a new role in auxin signaling for the CSN holocomplex or a distinct CSN3-containing complex. While deneddylation is the only known biochemical activity of the CSN itself, additional activities including deubiquitinylating and protein kinase activities have been reported to be associated with the CSN [44,65,66]. Furthermore, in animal systems some CSN subunits have been found to be DNA associated and suggested to regulate transcription [67,68]. Also, whether or not all CSN subunits function solely within the CSN holocomplex is unclear. On one hand, null mutations in any of the eight Arabidopsis subunits confer identical seedling-lethal phenotypes [23] and transcription profiles [35], suggesting that each subunit only functions within the CSN. Figure 6. A small CSN3-containing complex, but not the CSN, is affected by the csn3-3 mutation. Protein extracts from 7-d.o. Col, csn1-10 and csn3-3 seedlings were fractionated on a Superdex-200 gel filtration column and fractions (4 to 19) were collected and blotted with a-CSN1, a-CSN4, and a-CSN8 (A), a-CSN5 (B) and a-CSN3 (C). Fraction #4 was lost in the experiment examining CSN8. In (C), the assembly of a small CSN3containing protein complex was observed in fractions 12-14. This complex was absent in csn3-3 extracts, but was restored by introduction of a genomic CSN3 transgene (L4 complementation line). Molecular mass standards are shown at the bottom of each panel. EX indicates the protein crude extract before gel filtration. (D) CSN3 protein level is unaffected by the csn3-3 mutation. RPT5 is shown as a loading control. doi:10.1371/journal.pone.0066578.g006 However, it is possible that CSN subcomplexes or individual subunits have additional functions that are masked by the early seedling lethality of these null mutants. Consistent with this possibility, fission yeast csn1 and csn2 mutations confer DNA replication defects whereas other subunit mutations do not [25]. Similarly, while both Drosophila csn4 null and csn5 null mutants are embryo-lethal, these two mutants exhibit distinct developmental arrest phenotypes [63,69] and differentially affect gene expression [70].
Given that csn3-3 confers no apparent defects in cullin deneddylation, SCF TIR1/AFB activity, or CSN holocomplex assembly, but does specifically abolish the ,130 kD sCSN3c complex, we hypothesize that a defective sCSN3c may be the basis of the auxin signaling defects displayed by csn3-3 mutant plants ( Figure 7). Consistent with this possibility, expression of a P CSN3 :CSN3 transgene in csn3-3 mutant plants restored both the auxin response defects (Figures 1C, 2E) and the sCSN3c complex ( Figures 6C). Prior studies have reported CSN subunits in complexes smaller than the CSN holocomplex. While some support has been presented for CSN5 functioning autonomously of other CSN subunits [71,72], these smaller CSN complexes have generally been proposed to be mini-CSN complexes containing several, but not all subunits [64]. Whether these represent intermediates in CSN holocomplex assembly or functionally distinct complexes is uncertain. Interestingly, one study examining these mini-CSN complexes from animal cells by non-denaturing polyacrylamide electrophosesis detected CSN3-containing complexes that appeared to lack other CSN subunits [71]. In Arabidopsis, Rubio et al. [73] also detected, but did not discuss or characterize, small CSN3-containing complexes.
Recent analyses of CSN subunit interactions using in vitro reconstituted human CSN subunits suggest that the CSN consists of two symmetrical modules; CSN1/2/3/8 and CSN4/5/6/7 [64]. With this subunit topology, CSN3 would directly interact with CSN1/8 and possibly CSN4, which is consistent with prior two-hybrid studies [74]. However, our gel filtration analysis indicates that neither CSN1/4/8 nor CSN5 are components of the sCSN3c complex. While we cannot eliminate the possibility that CSN2/6/7 are sCSN3c components, this seems unlikely given these prior findings. Thus, together with our finding that CSN holocomplex assembly is unaffected by csn3-3, we hypothesize that sCSN3c represents a novel complex rather than a mini-CSN complex. Identification of the other components within sCSN3c may provide crucial information into what role this complex might play in auxin signaling. Furthermore, since the residue affected by the csn3-3 missense mutation is extremely highly conserved across eukaryotes, it seems likely that sCSN3c function may be similarly conserved.
For root growth assays, 5-d.o. seedlings were transferred to ATS medium supplemented with various concentrations of 2,4-D, and root growth was measured after an additional 4 days. Percent inhibition was calculated by dividing the average growth on auxin media by the average growth on ATS control media and subtracting this ratio from 100%. For measuring IAA-induced root growth inhibition, 6-d.o. seedlings were transferred to freshly made IAA plates and were grown under long-day illumination through yellow long-pass filters to slow indolic compound breakdown. Protein extractions for gel filtration and western experiments were made from 7 -10-d.o. seedlings grown in liquid ATS medium on a shaker at 20uC.

Complementation
The CSN3 genomic DNA construct was composed of the CSN3 coding region together with a 1.3 kb fragment upstream of the start codon plus a 600bp fragment downstream of the stop codon. The whole sequence was amplified from BAC clone F18022 (59-GGGGACAAGTTTGTACAAAAAAGCAGGCTCCTTTGAT GGCGCCATGGTGG-39 and 59-GGGGACCACTTTGTA-CAAGAAAGCTGGGTCGTATGGAAACATGTGATAACC) and Gateway cloned into the vector pEarleyGate301 [76]. The construct was then transformed into csn3-3[BA3:GUS] plants using Agrobacterium tumefaciens strain GV3101 according to standard procedures [77]. Two independent T3 homozygous lines (L2 and L4) were used for the root growth and GUS assays to assess complementation. Figure 7. The CSN3 subunit plays multiple roles in auxin signaling. We hypothesize that in addition to its role in the CSN as a cullin deneddylase, the CSN3 subunit also regulates auxin signaling independently of the SCF TIR1/AFB ubiquitin-ligase. This second regulatory mechanism may involve the small CSN3-containing complex whose assembly is disrupted by the csn3-3 point mutation. doi:10.1371/journal.pone.0066578.g007 GUS Histochemical Staining BA3:GUS and DR5:GUS assays were conducted as described previously with slight modifications [52]. 6-d.o. seedlings grown vertically on ATS plates were transferred into liquid ATS medium supplemented with 0.5 mM 2,4-D for 12 h (BA3:GUS) and 4 h (DR5:GUS) before GUS staining. For HS:AXR3NT-GUS assays, 6d.o. Col, csn1-10 and csn3-3 seedlings homozygous for the reporter construct were heat-shocked for 2 h at 37uC to induce expression of the transgene. Seedlings then were transferred to 20uC medium with 1 mM 2,4-D for 20 min before GUS detection. Quantitative measurements of b-glucuronidase activity were conducted as described previously [4] with modifications: after heat shock for 2 h at 37uC, seedlings were transferred to 20uC liquid ATS medium for 15 min. Seedlings were then moved into medium supplemented with 1 mM 2,4-D, sampled at 0, 20, 40 and 60 min thereafter and stored in liquid nitrogen until protein extraction. Frozen seedlings were homogenized by a laboratory vibration mill (Mixer-Mill; Qiagen, cat. no. MM300) in extraction buffer (50 mM KPO 4 pH 7, 0.1% triton X-100, 10 mM b-mercaptoethanol, 10 mM EDTA). Samples were then centrifuged to remove debris and the protein level of each sample was measured. Protein extracts were mixed with an equal volume of extraction buffer plus 4 mM MUG (4-methylumbelliferyl-B-D-glucoronide), incubated at 37uC overnight, and the fluorometric signal measured using a BioTek FL600 Fluorescence Microplate Reader (Winooski, VT) as per manufacturer's instructions (excitation at 360 nm, emission at 460 nm). Activity was calculated as fluorometric units per mg protein and were normalized as the percentage of the starting point (0 min).

IAA28-myc degradation assay
Experiments were done as described previously [60] with modifications. 7-d.o. light-grown seedlings of Col, csn1-10 and csn3-3 carrying the IAA28-myc construct were removed from ATS plates and floated in liquid ATS supplemented with 0.5 mM IAA. At the indicated time points, roots were excised and homogenized for protein extraction.

Gel Filtration Chromatography
Experiments were conducted as described previously [43]. In brief, 7-d.o. seedlings grown in liquid ATS medium were homogenized in extraction buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM EDTA, 10% glycerol, 10 mM MgCl 2 , 0.5 mM NaN 3 , 1 mM DTT, 1 mM PMSF, and 1X Protease Inhibitor Cocktail Kit (Thermo)). Homogenates were centrifuged twice for 10 min at 4uC to remove debris. Supernatants were then filtered though a 0.45 mm HT TuffrynH Membrane (Pall). 600 mg of total protein was fractionated through a Superdex 200 10/300 GL column (Amersham/GE Healthcare). After loading the sample, proteins were eluted in filtered and degassed extraction buffer at a flow rate of 0.2 mL/min. 0.5 mL fractions were collected after the 6 mL void volume was reached. All procedures were carried out at 4uC.

Supporting Information
Figure S1 IAA induced BA3:GUS expression is reduced by the csn1-10 and csn3-3 mutations to a similar extent. 6-d.o. transgenic Col, csn1-10 and csn3-3 seedlings carrying the BA3:GUS reporter were treated with 1 mM IAA for 3 h before histochemical staining for b-glucuronidase activity. (TIF) Figure S2 a-myc western detection of the IAA28-myc fusion protein. The P IAA28 :IAA28-myc reporter was introduced into the csn1-10 and csn3-3 backgrounds by crossing. Protein extracts were made from 7-d.o. seedling roots treated with IAA for the indicated time. RPT5 was used as a loading control. (TIF) Figure S3 sCSN3c is not a breakdown product of the CSN holocomplex during the in vitro fractionation. Fractions (#5,9) of the first gel filtration run using Col seedling protein extracts were isolated and injected into the column for a second round of gel filtration. CSN3 western detection was conducted using fractions from the 2nd gel filtration. No CSN3 was detected in the sCSN3c fractions. (TIF)