Regulation of retrotranslocation by p97-associated deubiquitinating enzyme ataxin-3

The endoplasmic reticulum (ER) is a major site of protein folding and assembly in eukaryotes. Polypeptides entering the ER often encounter various folding problems, resulting in aggregated or misfolded proteins. To preserve ER homeostasis, eukaryotes have evolved a conserved quality control pathway termed retrotranslocation, dislocation, or ER-associated protein degradation (ERAD), which efficiently eliminates misfolded ER proteins by exporting them into the cytosol for degradation by the ubiquitin–proteasome system (Meusser et al., 2005).

Retrotranslocation is initiated when misfolded polypeptides are selectively targeted to the site of translocation at the ER membrane from where they are subsequently dislocated into the cytosol. The export of a subset of substrates requires the Derlin proteins, members of a highly conserved multi-spanning membrane protein family postulated to form a channel (Lilley and Ploegh, 2004; Ye et al., 2004; Oda et al., 2006). Other substrates might use different routes to exit the ER. Most substrates undergoing retrotranslocation are modified by polyubiquitination, which is achieved by the sequential action of a ubiquitin-activating enzyme (E1), a ubiquitin-conjugating enzyme (E2), and a ubiquitin ligase (E3) (Meusser et al., 2005). Polyubiquitination occurs at the cytosolic side of the ER membrane after a portion of the substrate has emerged into the cytosol. This modification is required for the subsequent dislocation of polyubiquitinated substrates from the ER membrane, a process that is mediated by the p97–Ufd1–Npl4 ATPase complex (Bays and Hampton, 2002). During Derlin-dependent retrotranslocation, the ATPase complex is recruited to the Derlins together with a single spanning membrane protein called VIMP and certain ER-specific ubiquitin ligases (Ye et al., 2004, 2005; Lilley and Ploegh, 2005). The assembly of a retrotranslocation complex containing both the “pulling” ATPase p97 and the ubiquitination machinery ensures that substrate ubiquitination and dislocation are tightly coupled. For misfolded glycoproteins, their dislocation is also associated with the removal of the attached N-linked glycan by a cytosolic N-glycanase (Kim et al., 2006).

After entering the cytosol, polypeptides need to be transferred to the proteasome for degradation. This process is poorly defined. In Saccharomyces cerevisiae, the p97 homologue cdc48p does not seem to form a stable complex with the 26S proteasome (Verma et al., 2000). Substrates bound by p97 thus need to be shuttled to the proteasome, which likely involves interactions between ubiquitin conjugates and various ubiquitin binding proteins (Richly et al., 2005). Several proteasome-associated proteins such as rad23p in yeast contain ubiquitin binding motifs (Elsasser and Finley, 2005), and can serve as receptors to collect ubiquitinated substrates (Verma et al., 2004). However, little is known about the “shuttling factors” that deliver substrates to these receptors. One candidate for a shuttling factor is ataxin-3 (atx3) (Kuhlbrodt et al., 2005), a poly-glutamine (poly-Q) containing deubiquitinating enzyme (DUB) that interacts with both p97, and HHR23A and B, mammalian homologues of rad23p (Wang et al. 2000; Doss-Pepe et al., 2003; Matsumoto et al., 2004; Boeddrich et al., 2006). Expansion of the poly-Q segment in atx3 has been linked to a dominantly inherited form of spinocerebellar ataxia, a member of the poly-Q– induced family of neurodegenerative diseases (Kawaguchi et al., 1994), but how such poly-Q expansion causes the disease is unclear. In addition, although the enzymatic activity of atx3 has been extensively characterized (Burnett et al., 2003; Chow et al., 2004; Berke et al., 2005; Mao et al., 2005), its physiological function remains elusive. This study was initiated to explore the potential involvement of atx3 in ERAD. We conclude that atx3 acts in conjunction with p97 to regulate the degradation of misfolded ER proteins.

To test the potential involvement of atx3 in ERAD, we first characterized its interaction with p97 using purified components. Indeed, recombinant protein containing the GST fused to atx3 (GST-atx3) bound purified wild-type (wt) p97 as well as several p97 variants that were defective either in ATP binding (AA and KA) or in ATP hydrolysis (QQ) (Ye et al., 2003). In contrast, a p97 mutant lacking its N-terminal domain (N-domain) failed to bind GST-atx3 (Fig. 1 A). These results suggest that atx3 binds p97 via its N-domain. Because many cofactors bind the p97 N-domain in a mutually exclusive manner (Meyer et al., 2000), we tested whether atx3 would influence the interaction of p97 with Ufd1–Npl4 (U/N), a binary cofactor complex required for ERAD. Recombinant p97 was incubated with purified U/N in the presence of increased concentrations of either GST-atx3 or of a recombinant protein containing the GST fused to SVIP, a small p97 cofactor of unknown function (Nagahama et al., 2003). The p97–Ufd1–Npl4 complex immunoprecipitated with anti-Ufd1 antibodies was analyzed by immunoblotting. As previously reported (Nagahama et al., 2003), addition of SVIP resulted in reduced association of p97 with U/N, presumably because these cofactors compete for the same binding site on p97 (Fig. 1 B, lanes 5–7). In contrast, GST-atx3 did not affect the assembly of the p97–Ufd1–Npl4 complex (Fig. 1 B, lanes 1–4), suggesting that atx3 either uses a different mode to interact with the p97 N-domain or binds p97 with an affinity significantly lower than other cofactors. The lack of competition between atx3 and U/N raises the possibility that these cofactors may act in conjunction with p97 to regulate ERAD.

Therefore, we tested whether atx3 also interacts with other known components of the ERAD pathway, such as the recently identified Derlin–VIMP complex. Detergent extracts of 293T cells expressing FLAG-tagged atx3 were subjected to immunoprecipitation with anti-FLAG antibodies. Immunoblotting with specific antibodies showed that a fraction of Derlin-1, VIMP, and p97 was coprecipitated with atx3 (Fig. 2 A, lane 4). Endogenous atx3 could be coimmunoprecipitated with Derlin-1 and p97 from detergent extract of dog pancreatic ER membranes (Fig. 2 B, lane 3). When FLAG-tagged atx3 was coexpressed with Myc-tagged Hrd1, an ER-specific ubiquitin ligase (Bays et al., 2001), both proteins could be coimmunoprecipitated (Fig. 2 C). Together, these data suggest that a fraction of atx3 is associated with the ER membranes via interactions with components of the retrotranslocation machinery, including Derlin-1, VIMP, p97, and an ER-specific ubiquitin ligase. Because only a small fraction of the Derlin complex could be coprecipitated with atx3, and immunoprecipitation using the VIMP antibody pulled down little atx3 (see Fig. 5 A), atx3 appears to transiently interact with the ER membrane (see Discussion).

Because knock-down of atx3 by ∼80% using atx3-specific siRNA was not sufficient to cause an apparent defect in retrotranslocation (Fig. S1), we attempted to interfere with the function of endogenous atx3 with a catalytically inactive atx3 mutant (atx3 C14A). We reasoned that this atx3 variant was likely to exert a dominant-negative effect because it binds ubiquitinated substrates with high affinity, but cannot process them (Burnett et al., 2003). We first tested whether the expression of atx3 C14A inhibits the p97-dependent degradation of misfolded TCRα (Fig. S2) (DeLaBarre et al., 2006), a type I transmembrane protein targeted for retrotranslocation due to the lack of its assembly partners (Huppa and Ploegh, 1997). To this end, we expressed TCRα together with either FLAG-tagged wt atx3 or the atx3 C14A mutant in 293T cells. Quantitative immunoblotting using fluorescence-labeled secondary antibodies showed that the FLAG-tagged atx3 variants were expressed ∼12-fold higher than the endogenous protein (unpublished data). When the level of TCRα was determined, it was indeed elevated in cells expressing atx3 C14A (Fig. 3 A, lane 2 vs. lane 1; Fig. 3 B, lane 2 vs. lane 1). A pulse-chase experiment confirmed that TCRα was more stable in atx3 C14A–expressing cells (Fig. S3, t_1/2 ∼60 min in atx3 C14A cells vs. ∼30 min in control cells). Because a significant fraction of the stabilized TCRα appeared to carry polyubiquitin conjugates as determined by immunoblotting the precipitated TCRα with ubiquitin antibodies (Fig. 3 A, lane 2; Fig. 3 B, lane 2), the stabilization effect by atx3 C14A was likely underestimated in the pulse-chase experiment because only the nonubiquitinated TCRα species could be analyzed. In contrast to atx3 C14A, overexpression of wt atx3 only moderately stabilized TCRα (Fig. S3), and did not increase the level of ubiquitinated TCRα (Fig. 3 B, lane 3). atx3 C14A–induced accumulation of ubiquitinated TCRα was more apparent when cells were treated with the proteasome inhibitor MG132 (Fig. 3 A, lane 4). Although MG132-treated control cells also contained a significant amount of the ubiquitinated TCRα, most substrates in these cells remained nonubiquitinated, which included both the glycosylated retrotranslocation precursor molecules and the dislocated deglycosylated degradation intermediates (Fig. 3 A, lane 3). In contrast, atx3 C14A–expressing cells treated with MG132 accumulated more ubiquitinated TCRα molecules at a compensatory loss of the nonubiquitinated species (Fig. 3 A, lane 4 vs. lane 3). These data suggest that the deubiquitination of TCRα is inhibited in atx3 C14A–expressing cells, leading to the stabilization of TCRα in part as polyubiquitinated species.

To test for a direct involvement of atx3 in ERAD, we examined its interaction with TCRα. Detergent extracts of cells expressing HA-tagged TCRα together with either wt atx3 or atx3 C14A were subjected to immunoprecipitation with HA antibodies. Immunoblotting analysis showed that only a small fraction of wt atx3 was coprecipitated with TCRα (Fig. 3 C, lane 3), whereas significantly more atx3 C14A was bound to TCRα (Fig. 3 C, lane 2). These results are consistent with the concept that the atx3 C14A mutant is able to sequester its substrates in a stable complex. Treatment of cells with MG132 stabilized TCRα (Fig. 3 C, lanes 4–6 vs. lanes 1–3), and further enhanced the interaction between TCRα and atx3 C14A (Fig. 3 C, lane 5 vs. lane 2). In contrast, the association of TCRα with wt atx3 remained weak (Fig. 3 C, lane 6). These data suggest that atx3 transiently binds TCRα to directly regulate its degradation.

To see whether atx3 also plays a role in degradation of soluble ERAD substrates, we examined the degradation of misfolded β-site amyloid precursor protein cleaving enzyme (BACE457Δ) (Molinari et al., 2003) in cells expressing either wt atx3 or atx3 C14A. Similar to TCRα, BACE457Δ was stabilized in atx3 C13A–expressing cells, and a significant fraction of the accumulated BACE457Δ contained polyubiquitin conjugates (Fig. 3 D, lane 3). In contrast, expression of wt atx3 only moderately increased the level of nonubiquitinated BACE457Δ (Fig. 3 D, lane 2 vs. lane 1). Thus, atx3 is also involved in the degradation of lumenal ERAD substrates.

Because a defect in ERAD is usually associated with the accumulation of misfolded proteins in the ER, which triggers ER stress, we monitored the ER stress level in cells expressing atx3 C14A to further confirm its involvement in ERAD. Indeed, expression of atx3 C14A caused strong induction of an ER stress reporter gene, pGL3-GRP78 (-132)-luciferase (Fig. 3 E). Expression of wt atx3 also induced ER stress, albeit to a much lesser degree. This is consistent with the observation that wt atx3 caused a moderate stabilization of TCRα and BACE457Δ. Together, these results further support the notion that the deubiquitinating activity of atx3 may be essential for efficient elimination of misfolded ER proteins.

The inhibition of ERAD by atx3 C14A cannot be attributed to general perturbation of the ubiquitin proteasome system by the overexpressed defective DUB, because the proteasome-dependent degradation of an N-end rule substrate Ub-R-GFP (Dantuma et al., 2000) was not affected (Fig. 3 F). Interestingly, the degradation of Ub-R-GFP could be inhibited by a p97 mutant that was defective in ATP hydrolysis (QQ). Thus, the inhibitory effect of atx3 C14A on protein degradation cannot be simply due to nonselective disruption of p97 function by the overexpressed mutant protein.

Because atx3 is a p97-associated DUB (Burnett et al., 2003), it may facilitate ERAD by promoting deubiquitination of substrates bound by p97. To examine p97-associated deubiquitination (PAD), p97 and its associated proteins were immunoprecipitated from cell extracts and analyzed by immunoblotting. A fraction of the precipitated material carried polyubiquitin conjugates. Because these ubiquitinated proteins were not detected when immunoprecipitation was performed under denaturing condition, they were likely substrates bound to p97 (Fig. S4). Next, we tested whether these ubiquitinated proteins could be subjected to deubiquitination in vitro by DUBs coprecipitated with p97. Indeed, the amount of p97-bound ubiquitinated material was slowly reduced during incubation (Fig. 4 A, lanes 1–4), a process that could be blocked by N-ethylmaleimide, an inhibitor of deubiquitination (unpublished data). To test whether atx3 can facilitate this process, p97 and its associated substrates purified from cells expressing either wt atx3 or atx3 C14A mutant were tested for in vitro deubiquitination. Expression of wild type, but not the catalytically inactive atx3 mutant, accelerated the decay of these ubiquitin conjugates (Fig. 4 A, lanes 5–8 vs. lanes 1–4; Fig. 4 B; Fig. S5). These data indicate that atx3 indeed facilitates PAD. Notably, a smaller amount of ubiquitinated proteins was coprecipitated with p97 from cells expressing wt atx3 (Fig. 4 A, lane 5 vs. lane 1; Fig. 4 C, lane 3 vs. lane 1), suggesting that atx3-mediated PAD may also occur in intact cells. Expression of USP14, a DUB primarily associated with the proteasome (Borodovsky et al., 2001; Leggett et al., 2002), did not decrease the amount of ubiquitinated proteins bound to p97 (Fig. 4 C, lane 5 vs. lane 1), and did not increase the rate of PAD in vitro (Fig. 4 C, lanes 5 and 6). Together, these data suggest that atx3 can promote PAD both in intact cells and in vitro.

To understand how atx3 cooperates with p97 to promote deubiquitination, we wished to determine the order between atx3-mediated deubiquitination and p97-dependent ATP hydrolysis. We reasoned if atx3 acts subsequent to the ATPase cycle of p97, inhibition of p97 ATPase activity would likely prevent ubiquitinated substrates from reaching the enzymatic site of atx3 because substrates cannot be released from p97 under this condition (Ye et al., 2001). To test this idea, FLAG-tagged wt atx3 was coexpressed together with either His-tagged wt p97 or a p97 mutant defective in ATP hydrolysis (QQ) (Ye et al., 2001). The amount of ubiquitinated substrates associated with His-tagged p97 was determined. The expression of atx3 reduced the amount of ubiquitinated proteins associated with wt p97 (Fig. 4 D, lane 2 vs. lane 1), but not of those bound by p97 QQ (lane 4 vs. lane 3), although more atx3 was bound by p97 QQ (lane 4 vs. lane 2). Moreover, the expression of p97 QQ resulted in more ubiquitinated proteins in association with atx3 and p97 (Fig. 4 E, lane 5 vs. lane 1), and these ubiquitinated proteins were more resistant to deubiquitination in vitro (Fig. 4 E, lanes 5–8 vs. lanes 1–4; Fig. 4 F). These data indicate that ubiquitinated substrates accumulated upon p97 QQ expression cannot be efficiently deubiquitinated by atx3. Although these substrates can be coimmunoprecipitated with atx3, they are likely still bound by the p97 QQ mutant, which is also present in the complex. Thus, ATP hydrolysis by p97 appears to precede atx3-mediated deubiquitination. Perhaps p97-dependent ATP hydrolysis is coupled to the transfer of substrates to the enzymatic site of atx3. Because the p97 ATPase cycle is required for the extraction of misfolded proteins from the ER membranes (Ye et al., 2001), atx3 likely acts at a step downstream of dislocation.

We noticed that more atx3 was bound to the p97 QQ than to wt p97 (Fig. 4 D, lane 4 vs. 2; Fig. 4 E). In addition, an increased association between atx3 and VIMP was observed in cells expressing p97 QQ (Fig. 5 A, lane 3 vs. lanes 1 and 2). Consequently, more atx3 was found in association with the ER membrane in these cells (Fig. 5 B, lane 5 vs. lanes 1 and 3). These data indicate that the association of atx3 with p97 and VIMP at the ER membrane may be regulated by the p97 ATPase cycle, a conclusion that is also supported by the recent observation that the interaction of p97 with atx3 in brain extracts is enhanced when ATP or ATPγS is present (Boeddrich et al., 2006). Because atx3 bound purified wt p97 and the p97 QQ mutant with similar affinity in vitro (Fig. 1 A), the enhanced interaction between atx3 and p97 QQ mutant may require additional factors present in cells. Interestingly, mutant atx3 lacking its deubiquitinating activity (C14A) also exhibited an increased affinity to p97, VIMP, and Derlin-1 (Fig. S5; Fig. 5 C). Consistent with this finding, atx3 C14A also accumulated at the ER membrane to a much higher level than wt atx3 (Fig. 5 D, lane 3 vs. lane 1). The simplest explanation of these observations is that a small fraction of atx3 is transiently associated with the ER membrane via interactions with p97 and VIMP. atx3 may shuttle on and off p97 to promote PAD, a process that is linked to the ATPase cycle of p97 and to the subsequent delivery of substrates to the proteasome. Mutations affecting the enzymatic activity of either p97 or atx3 appear to interfere with the release of atx3 from p97, which may also prevent substrates from leaving p97 (see following paragraph).

To test whether atx3 C14A inhibits ERAD by interfering with the transfer of substrates from p97 to the proteasome, we determined the relative amount of ubiquitinated substrates bound to p97 versus those bound by the proteasome. We reasoned that a defect in substrate transfer would lead to an accumulation of ubiquitinated substrates in complex with p97. Detergent extracts of 293T cells were subjected to immunoprecipitation with antibodies to either p97 or α6, a subunit of the 20S proteasome. Immunoblotting showed that the α6 antibodies precipitated α7, another subunit of the 20S proteasome, together with Mms1, a subunit of the 19S particle, but not p97. Likewise, antibodies against p97 precipitated the ATPase itself, some Mms1, and no α7 (Fig. 6 A). These data confirm that p97 and the 26S proteasome do not form a stable complex. When ubiquitinated proteins were examined, control cells contained little polyubiquitinated proteins, and they were evenly distributed between the proteasome and p97 (Fig. 6 A, lane 2 vs. lane 1). In contrast, atx3 C14A–expressing cells contained more ubiquitinated proteins, and they were almost entirely associated with p97 (Fig. 6 A, lanes 3 and 4; Fig. 6 B). The increased association of p97 with ubiquitinated proteins cannot be simply due to the accumulation of ubiquitinated proteins in cells expressing a defective DUB, because overexpression of the catalytically inactive USP14 mutant (USP14 C114A) also increased the level of ubiquitinated proteins, but did not cause an apparent accumulation of ubiquitinated proteins in complex with p97 (Fig. 6 B). These data indicate that atx3 C14A likely blocks the transfer of substrates from p97 to the proteasome, leading to their stabilization.

We have investigated the role of atx3 in retrotranslocation of misfolded ER proteins. We show that atx3 interacts with the recently identified retrotranslocation complex consisting of Derlin-1, VIMP, certain ER-specific ubiquitin ligases, and the pulling ATPase p97 in the ER membrane. A direct involvement of atx3 in ERAD is suggested by its binding to model ERAD substrates en route to the proteasome. Moreover, expression of a mutant atx3 defective in deubiquitination (atx3 C14A) dramatically inhibits the degradation of both membrane and lumenal ERAD substrates. Consequently, ER stress is induced in these cells. ERAD substrates stabilized by mutant atx3 are accumulated in part as polyubiquitinated degradation intermediates, suggesting that atx3 acts on these substrates to remove their ubiquitin conjugates. Intriguingly, ubiquitinated proteins accumulated upon mutant atx3 expression are sequestered in complex with p97, raising the possibility that atx3-mediated deubiquitination may be coupled to the transfer of substrates from p97 to the proteasome.

The observation that overexpression of wt atx3 moderately inhibits ERAD is unexpected, but is not without precedence. In fact, overexpression of rad23, an atx3-interacting partner that facilitates the delivery of ubiquitinated proteins to the proteasome (Elsasser and Finley, 2005), also inhibits protein turn-over. Recent studies suggest that the effect of rad23 on protein degradation is highly dependent on its concentration, with low concentrations being stimulatory and high concentrations inhibitory (Verma et al., 2004). The inhibition of ERAD by wt atx3 may be similarly caused by its overexpression, which may titrate out certain factors essential for ERAD. Alternatively, the deubiquitinating activity of atx3 may need to be tightly regulated in conjunction with other reactions to mediate the degradation of ERAD substrates (see following paragraph). The presence of excessive atx3 may uncouple deubiquitination from these processes, which may cause a defect in ERAD.

We propose that atx3 may promote deubiquitination of p97-bound substrates to facilitate their transfer to the proteasome during retrotranslocation (Fig. 7). In this model, atx3 would first interact with p97, which is prebound to its substrates together with Derlin-1 and VIMP at the ER membrane. In the next step, substrates may be transferred from p97 to atx3, a process that is likely associated with the p97 ATPase cycle. Next, atx3 would act on the ubiquitinated substrates, which might be subsequently transferred to a downstream ubiquitin receptor such as rad23. Because ubiquitin chains consisting of at least four ubiquitin molecules are required for efficient targeting of modified substrates to the proteasome, deubiquitination of substrates en route to the proteasome by atx3 is likely tightly coupled to the transfer process to ensure that the targeting signal is not prematurely erased. In S. cerevisiae, ubiquitin chains on substrates of cdc48p appear to be progressively shortened while the substrates are in transit to the proteasome (Richly et al., 2005). In light of this finding, we speculate that atx3 may function as an editing enzyme to trim ubiquitin conjugates on substrates to guide them to the proteasome. Although atx3 immunoprecipitated from mammalian cells appears to deubiquitinate its substrates completely, recombinant atx3 purified from bacteria indeed exhibits chain trimming activity, leading to a shortening of ubiquitin chains rather than to their complete removal (Burnett, et al., 2003). Because atx3 binds the ubiquitin-like domain of rad23, which also mediates the interaction between rad23 and the proteasome (Doss-Pepe et al., 2003), it is likely that rad23 may not interact with atx3 and the proteasome simultaneously.

Our analysis demonstrates a physiological function for atx3. We postulate that the expression of the poly-Q–expanded atx3 mutants may also impair ERAD and trigger ER stress, which may contribute to the pathogenesis of spinocerebellar ataxia. Further analysis of these mutants will help to clarify the long-suspected connection between ER stress and neurodegenerative diseases.

The plasmids pcDNA3.1 His-p97 wt and pcDNA3.1 His-p97 QQ were described previously (Ye et al., 2004). The plasmid pRK-FLAG atx3 wt (consisting of 20 CAG repeat) was constructed by cloning the human atx3 cDNA (a gift from R. Pittman, University of Pennsylvania, Philadelphia, PA; Burnett et al., 2005) into the SalI and NotI sites of the pRK-FLAG vector (a gift from H. Shu, the National Jewish Medical Center, Denver, CO). pRK-FLAG-USP14 was constructed by cloning the human USP14 cDNA into the SalI and NotI sites of the pRK-FLAG vector. pRK-FLAG-atx3 C14A and pRK-FLAG-USP14 C114A were constructed by site-directed mutagenesis using the QuikChange mutagenesis kit from Stratagene. All plasmids were sequenced. HA-tagged TCRα clone was a gift from R. Kopito (Stanford University, Stanford, CA; Yu et al., 1997). pGL-3-GRP78 (-132) luciferase plasmid was provided by K. Mori (Kyoto University, Kyoto, Japan). Plasmids expressing Ub-R-GFP, GST-SVIP, and BACE457Δ were provided by N. Dantuma (Karolinska Institute, Stockholm, Sweden), M. Nagahama (Tokyo University of Pharmacy and Life Science, Tokyo, Japan), and M. Molinari (Institute for Research in Biomedicine, Bellinzona, Switzerland), respectively.

Anti-BACE antibody was a gift from P. Paganetti (Novartis Pharma AG, Basel, Switzerland). Antibodies to ubiquitin, p97, Ufd1, and VIMP were described previously (Ye et al., 2004, 2005). Rabbit polyclonal antibodies to Derlin-1 were raised against a peptide corresponding to residues 238–251 of the human Derlin-1. All antibodies used in this study were affinity purified using the corresponding antigenic peptides. FLAG, HA, and GFP antibodies were purchased from Sigma-Aldrich, Roche, and Invitrogen, respectively. MG132 was purchased from Calbiochem.

The purification of His-tagged p97 variants and the Ufd1–Npl4 complex has been described previously (Ye et al., 2003). GST-atx3 was purified according to standard procedure using glutathione beads. GST pull-down experiments were performed as described previously (Ye et al., 2005). In brief, GST fusion proteins (5 μg) bound to glutathione beads were incubated with various purified recombinant proteins in 50 mM Hepes (pH 7.3), 150 mM potassium chloride, 2.5 mM magnesium chloride, 5% glycerol, and 0.1% Triton X-100. Bound material was analyzed by SDS/PAGE and stained with Coomassie blue.

293T cells were maintained according to standard procedures. Transfections in 293T cells were done with Trans IT (Mirus). For fractionation experiments, cells were homogenized by passing through a 25G needle in a low salt buffer (50 mM Hepes, pH 7.4, 10 mM potassium chloride, 1 mM DTT, and a protease inhibitor). Cell extracts were then subjected to sequential centrifugation at different speeds. For coimmunoprecipitation experiments, tissue culture cells or dog pancreatic microsomes were extracted with buffer N (30 mM Tris/HCl, pH 7.4, 150 mM potassium acetate, 4 mM magnesium acetate, 1% DeoxyBigCHAP, and a protease inhibitor cocktail). To detect ubiquitinated TCRα, cell extracts were adjusted to contain 0.2% SDS, 0.5 mM DTT, and 5 mM N-ethylmaleimide. Samples were heated at 95°C before being subjected to immunoprecipitation. To isolate the 26S proteasome, 2 mM ATP was included in the extraction buffer to preserve the interaction between the 19S and the 20S proteasome particles. All cell extracts were cleared by centrifugation (20,000 g) and subjected to immunoprecipitation with various antibodies. Immunoblots were visualized with a cooled CCD digital camera system (LAS-3000; Fuji). Results were quantified using MultiGauge v3.0 software (Fuji).

To detect p97 or atx3-associated deubiquitination, cells were extracted with buffer N. Cell extracts were subjected to immunoprecipitation with either p97 antibodies or anti-FLAG antibody to isolate p97 or FLAG-tagged atx3. Proteins bound to the beads were incubated in a deubiquitinating buffer (50mM Tris/HCl, pH 7.4, 20mM potassium chloride, 5mM magnesium chloride, 1mM DTT, 2.5% BSA) at 37°C. Deubiquitination was stopped by addition of SDS sample buffer. Samples were analyzed by immunoblotting.

Fig. S1 shows that knock-down of atx3 by ∼80% by siRNA has little effect on the degradation of TCRα. Fig. S2 shows that the degradation of TCRα requires p97. Fig. S3 provides evidence that the accumulation of TCRα in atx3 C14A–expressing cells is due to its stabilization. Fig. S4 shows the association of p97 with ubiquitinated proteins. Fig. S5 demonstrates that an atx3 mutant defective in deubiquitination fails to promote PAD.

We thank W. Yang, H. Bernstein, T. Rapoport, and M. Gellert for critical reading of the manuscript; R. Kopito, K. Mori, H. Shu, R. Pittman, N.P. Dantuma, M. Nagahama, M., Molinari, and P. Paganetti for reagents; and R. Pittman for exchanging unpublished results. We are indebted to T. Rapoport for his continuous support and encouragement.

This research was supported by the Intramural Research Program of the National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases.

Note added in proof. While this manuscript was under review, a paper describing related findings appeared: Zhong, X., and R.N. Pittman. 2006. Hum. Mol. Genet. 15:2409–2420.