The ubiquitin-like modifier FAT10 interacts with HDAC6 and localizes to aggresomes under proteasome inhibition

The ubiquitin-proteasome system (UPS) is responsible for the targeted destruction of the majority of intracellular proteins. Proteasomal degradation is a tightly controlled process and can serve either of the following two functions in the cell: it can have a regulatory role by destroying and thus inactivating specific proteins such as cell-cycle regulators, key enzymes and transcription factors (Hershko and Ciechanover, 1998), or it can ensure the removal of non-functional, damaged or misfolded proteins from the cell (Goldberg, 2003). Substrates are targeted to the 26S proteasome by covalent modification with Lys48-linked polyubiquitin chains. Enzymatic cascades involving a ubiquitin-activating enzyme (E1), a ubiquitin-conjugating enzyme (E2) and a ubiquitin-ligase (E3) catalyze the formation of an isopeptide bond between the ϵ-amino group of a lysine residue in the target protein and the C-terminal diglycine motif of ubiquitin. Specificity in ubiquitin conjugation is conferred by a multitude of different E3 enzymes, which recruit one of a limited number of E2 enzymes, and the target protein, to facilitate attachment of the polyubiquitin chain (Hershko and Ciechanover, 1998; Pickart and Fushman, 2004). Although Lys48-linked polyubiquitylation is the main route of targeting for proteasomal degradation, there are a few exceptions to this rule. These include the ubiquitin-independent degradation of select proteins, such as ornithine decarboxylase (Murakami et al., 1992) or p21 (Chen et al., 2004), as well as the attachment of polyubiquitin chains of unconventional linkage, such as Lys11- or Lys29-linked chains (Pickart and Fushman, 2004), or modification with the ubiquitin-like modifier FAT10 (also known as UBD) (Hipp et al., 2005).

Under normal conditions, the proteasome is responsible for the destruction of the majority of misfolded proteins in the cell. However, when the capacity of the proteasome is exceeded, the resulting accumulation of aggregation-prone misfolded proteins tends to have deleterious effects on the cell, which might contribute to a variety of conformational diseases such as Parkinson disease (PD), amyotrophic lateral sclerosis or dementia with Lewy bodies (DLB) (Gregersen et al., 2006). Over the past few years, evidence has been uncovered for a compensatory mechanism by which, under conditions in which proteasomal degradation is no longer possible, misfolded proteins are transported to pericentriolar inclusions termed aggresomes via dynein-dependent retrograde transport along the microtubule network (Johnston et al., 1998; Johnston et al., 2002; Garcia-Mata et al., 2002). Once sequestered in the aggresome, these proteins can be removed by the lysosomal pathway via autophagy (Iwata et al., 2005; Pandey et al., 2007). Making the connection between the UPS and aggresome formation is one of the proposed functions of histone deacetylase 6 (HDAC6), which, in addition to two catalytic domains, contains a ubiquitin-binding zinc finger (BUZ domain, also known as the PAZ, ZnF-UBP or DAUP domain) and a dynein-binding domain. The latter two domains are used by HDAC6 to function as a linker between the dynein motor complex and polyubiquitylated cargo during its transport to the aggresome (Hook et al., 2002; Seigneurin-Berny et al., 2001; Kawaguchi et al., 2003). In addition to its function as a linker, HDAC6 is also involved in aggresome formation and protein quality control in other ways – it mediates microtubule stability by the deacetylation of α-tubulin (Hubbert et al., 2002), it is involved in the transport of components of the autophagy apparatus to the aggresome (Iwata et al., 2005) and it is required for the induction of major cellular chaperones via HSF1 after misfolded-protein stress (Boyault et al., 2007). In addition, it can also deliver proteins to the aggresome directly, even under conditions in which the proteasome is functioning normally, if they are marked with a Lys63-linked polyubiquitin chain (Olzmann et al., 2007).

In addition to ubiquitin, the ubiquitin-like modifier FAT10 can serve as a signal for proteasomal degradation (Kerscher et al., 2006). Similar to ubiquitin, FAT10 can become covalently conjugated to its target proteins via a free diglycine motif at its C-terminus (Raasi et al., 2001; Chiu et al., 2007). It is encoded in the major histocompatibility class I locus, encompasses two ubiquitin-like domains (UBLs) connected by a short linker (Fan et al., 1996) and can be induced with the proinflammatory cytokines IFN-γ and TNF-α (Raasi et al., 1999; Liu et al., 1999). Overexpression of FAT10 leads to apoptosis in a conjugation-dependent manner (Raasi et al., 2001). Attachment of FAT10 causes the rapid degradation of long-lived proteins, which is dependent on the 26S proteasome but occurs independently of ubiquitylation (Hipp et al., 2005). In contrast to ubiquitin, FAT10 has a relatively short half-life because it is also subject to proteasomal degradation in its monomeric form and, additionally, it is probably not recycled but instead degraded along with its substrates (Hipp et al., 2004).

The results presented herein uncover a role for HDAC6 not only in the transport of polyubiquitylated cargo, but also in the transport of the ubiquitin-like modifier FAT10 and a FAT10-linked protein. We show that, after inhibition of proteasome activity, FAT10 interacts with HDAC6 and localizes to the aggresome in a microtubule-dependent manner. This interaction is mediated by the ubiquitin-binding zinc finger and the first catalytic domain of HDAC6 as well as both ubiquitin-like domains of FAT10, and is not dependent on catalytic activity of HDAC6. Combined with previous findings, these results suggest that the role of FAT10 is the rapid and inducible destruction of its target proteins via conjugation and subsequent degradation. If the primary way of degradation by the proteasome fails, an alternative route is taken to ensure their removal by sequestering them in the aggresome.

A yeast two-hybrid screen using FAT10 as a bait against a human lymph-node cDNA library (Hipp et al., 2004) identified HDAC6 as a novel interaction partner of FAT10. Sequencing of the plasmid recovered from the transformant revealed that the insert did not encode a full-length copy of HDAC6 but rather an N-terminal truncation containing only the last 106 amino acids of the protein. This C-terminal part of HDAC6 encompasses little more than the zinc-finger domain (BUZ), which is responsible for the interaction with free ubiquitin and ubiquitin conjugates (Seigneurin-Berny et al., 2001; Hook et al., 2002). To confirm the interaction with FAT10 in vivo, we transfected HEK293T cells with FLAG-tagged HDAC6 as well as hemagglutinin (HA)-tagged FAT10 and treated them with proteasome inhibitor before performing co-immunoprecipitation experiments. In accordance with the finding that HDAC6 only associated non-covalently with polyubiquitylated proteins if the proteasome is impaired (Kawaguchi et al., 2003), we observed robust interaction of FAT10 with wild-type HDAC6 only after proteasome inhibition (Fig. 1A, lanes 1 and 2). As shown in Fig. 1B, transiently transfected HDAC6 was also capable of interacting with endogenous FAT10 after proteasome inhibition. To our surprise, deletion of the BUZ domain did not abolish the interaction between FAT10 and HDAC6 (Fig. 1A, lanes 3 and 4), suggesting that – although being able to bind FAT10 on its own as shown by the yeast two-hybrid screen – the BUZ domain is not the only domain of HDAC6 that is capable of binding to FAT10.

Since overexpression of the other interaction partner discovered in the yeast two-hybrid screen, the ubiquitin-domain protein NEDD8 ultimate buster-1 long (NUB1L), leads to a marked acceleration of FAT10 degradation (Hipp et al., 2004), and because HDAC6 is able to attenuate the proteasomal degradation of polyubiquitylated misfolded proteins (Boyault et al., 2006), we decided to investigate whether overexpression of HDAC6 had an effect on the rate of FAT10 degradation. We performed pulse-chase assays with FAT10 in the presence or absence of HDAC6; however, as can be seen in Fig. 1C, coexpression of HDAC6 did not affect the degradation of FAT10.

To identify additional FAT10-binding domains of HDAC6, we performed co-immunoprecipitation experiments with truncation mutants of HDAC6 (Fig. 2A). To this aim, cells were transfected with both FAT10 and, individually, each of the HDAC6 mutants depicted in Fig. 2B, and treated with proteasome inhibitor. In these experiments, we were able to identify the first deacetylase (CAT1) domain of HDAC6 as a second domain responsible for interacting with FAT10. Interestingly, this is not the domain that exhibits α-tubulin deacetylase activity (Haggarty et al., 2003). Both the isolated BUZ as well as the CAT1 domain were able to associate with FAT10 after proteasome inhibition, but not as strongly as the full-length protein. However, the longer proteins that lack only the N-terminal (ΔN) or C-terminal (ΔBUZ) domain were able to precipitate amounts of FAT10 that were comparable to those pulled down by wild-type HDAC6. The 1-503 mutant showed weak interaction with FAT10 only in some of the experiments performed, whereas the 1-840 mutant was never able to bind FAT10, which might be explained by incorrect folding or the formation of multimers, which could result in blockage of the FAT10-binding site.

To confirm the results obtained in cell lines in an in vitro setting, we performed a GST-pulldown assay using recombinant GST-FAT10 and in vitro transcribed and translated [³⁵S]-methionine-labeled HDAC6. As shown in Fig. 3A, only GST-FAT10, but not GST alone, was able to pull down the CAT1 and the BUZ domains, and full-length HDAC6. Both of the isolated domains showed robust interaction with FAT10, whereas the isolated BUZ domain only interacted weakly in the co-immunoprecipitation experiments. Both in vitro and in intact cells, the full-length protein displayed the highest affinity for FAT10. Taken together, these results demonstrate that both the BUZ and the CAT1 domain are sufficient to bind FAT10 on their own, but do so much more efficiently in the context of the full-length protein.

Previous studies have suggested that the interaction between HDAC6 and ubiquitin is largely mediated by the free C-terminal diglycine of ubiquitin (Pai et al., 2007; Reyes-Turcu et al., 2006); therefore, we decided to investigate whether the C-terminal diglycine motif of FAT10 was required for its interaction with HDAC6. As demonstrated in Fig. 3C, both wild-type FAT10 and a mutant lacking the C-terminal diglycine were able to pull down comparable amounts of radiolabeled HDAC6. Further evidence that the C-terminal diglycine of FAT10 is not required comes from experiments in which a linear fusion of FAT10 to GFP was transiently transfected into HEK293T cells. In this fusion protein, the two C-terminal glycines of FAT10 were exchanged for alanine and valine (AV), yet it was still able to localize to aggresomes and could still be co-precipitated with HDAC6.

The role of the deacetylase activity of HDAC6 in aggresome formation remains somewhat mysterious. On one hand, the reintroduction of either ΔBUZ HDAC6 or a catalytically inactive mutant into HDAC6 knock-down cells resulted in little to no rescue of aggresome formation. On the other hand, formation of ubiquitin-free aggresomes, which appear after expression of the aggregation-prone GFP chimera GFP-250, which is not ubiquitylated (Garcia-Mata et al., 1999), does not seem to require HDAC6 at all (Kawaguchi et al., 2003). Then again, the inhibition of tubulin deacetylase activity has a profound effect on the transport of LC3, a component of the autophagy apparatus, to the aggresome, but affects the transport of polyubiquitylated misfolded huntingtin to a much lesser extent (Iwata et al., 2005). We therefore decided to investigate whether the catalytic activity of HDAC6 was required for its interaction with FAT10. In fact, this turned out not to be the case. A catalytically dead mutant, in which the deacetylase activity was inactivated by two point mutations in the active sites, interacted with FAT10 just as well as the catalytically active wild-type HDAC6 (Fig. 4A). Treatment of cells with the broad-spectrum deacetylase inhibitor, trichostatin A (TSA), even led to an increased association of HDAC6 with FAT10; this association could already be observed in the absence of proteasome inhibition (compare Fig. 4B, lanes 1 and 2 vs 5 and 6). To elucidate whether this increase was a direct effect of HDAC6 inhibition or a secondary consequence of inhibiting one or more of the many deacetylases also affected by TSA, we repeated the experiment using the HDAC6-specific tubulin deacetylase inhibitor tubacin (Haggarty et al., 2003). Treatment of cells with tubacin had hardly any effect on the association of FAT10 with HDAC6, although it markedly increased the level of acetylated α-tubulin (Fig. 4C). The addition of neither inhibitor affected HDAC6-FAT10 interaction in in vitro pulldown experiments (supplementary material Figs S1, S2), suggesting that the inhibition of catalytic activity has no influence on the binding capacity of HDAC6 for FAT10. We therefore conclude that the deacetylase activity of HDAC6 is dispensable for its interaction with FAT10.

The findings that: (1) HDAC6 interacts with FAT10 through one of the catalytic domains of HDAC6, (2) treatment with TSA slightly increases the interaction between the two proteins and (3) two-dimensional (2D) gel electrophoresis of FAT10 reveals several closely migrating bands with a similar molecular weight but distinct isoelectric points (Raasi et al., 2001) collectively raise the questions, is FAT10 acetylated and is this a prerequisite for its interaction with HDAC6? To address these questions, we investigated whether immunoprecipitated FAT10 showed reactivity with an antibody against acetylated lysine. As can be seen in supplementary material Fig. S3A, FAT10 can indeed be acetylated, although this is in all likelihood not a very dynamic process, as treatment of cells with TSA for 6 hours did not change the level of FAT10 acetylation. In addition, FAT10 does not appear to be a substrate of HDAC6 either, because FAT10 exhibited similar levels of acetylation when expressed in the presence or absence of HDAC6 (supplementary material Fig. S3B). The degree of FAT10 acetylation is estimated to be low, as judged from 2D gels (Raasi et al., 2001) and because of the fact that exposure times for the anti-acetylated-lysine western blots (Fig. 4E; supplementary material Fig. S3) were at least tenfold longer than for the anti-HA blots. In fact, the pool of FAT10 that interacted with HDAC6 after proteasome inhibition did not appear to be acetylated at all, as indicated by the complete absence of anti-acetyl-lysine reactivity (Fig. 4E). Moreover, a mutant of FAT10 in which all lysines were mutated to arginine (Hipp et al., 2005) – thus preventing it from being ubiquitylated or acetylated – was still able to interact with HDAC6 as well as, if not better than, wild-type FAT10 (Fig. 4D). Still, the ability of catalytically dead or inhibited HDAC6 to interact with FAT10 might solely be mediated through the C-terminal BUZ domain. To exclude this possibility, we performed co-immunoprecipitation experiments with an isolated, catalytically dead CAT1 domain. As can be seen in supplementary material Fig. S1A, the mutated CAT1 domain was still able to pull down FAT10. In addition, the isolated CAT1 domain was still capable of interacting with the lysine-less mutant of FAT10. supplementary material Fig. S1B demonstrates that recombinant GST-FAT10 was able to pull down similar amounts of the wild-type and mutated in vitro transcribed and translated [³⁵S]-methionine-labeled CAT1 domains. Together, these findings suggest that, although a small portion of the intracellular FAT10 is acetylated, acetylation is not required for its binding to HDAC6, nor is FAT10 a substrate of HDAC6 deacetylase activity.

As FAT10 interacts with HDAC6 after proteasome inhibition, and as HDAC6 is involved in the transport of polyubiquitylated proteins to the aggresome upon proteasome impairment (Kawaguchi et al., 2003), we investigated whether HDAC6 might likewise transport FAT10 to the aggresome. To test this hypothesis, we investigated whether treatment of transiently HA-FAT10-tranfected HEK293T cells with proteasome inhibitor had an influence on the subcellular localization of FAT10 as analyzed by confocal immunofluorescence microscopy. In untreated cells, FAT10 was evenly distributed throughout the cytoplasm and showed varying degrees of localization to the nucleus (Fig. 5Aa,i). HDAC6, by contrast, was completely excluded from the nucleus and showed only cytoplasmic localization (Fig. 5Ab). As reported by Kawaguchi et al. (Kawaguchi et al., 2003), proteasome inhibition induced the relocalization of HDAC6 to a single prominent juxtanuclear structure – and indeed also caused FAT10 to localize to the very same structure (Fig. 5Ae-h). To determine the identity of the observed inclusion bodies, we analyzed them for the presence of known aggresome markers other than HDAC6. We never observed formation of more than one of these inclusion bodies per cell and, combined with the observation that they colocalized with γ-tubulin (Fig. 5Am-p), a component of centromeres, this leads us to infer that they form at the microtubule organizing center. In addition, we observed a characteristic distortion and enlargement of the centromeres upon proteasome inhibition (compare Fig. 5Aj and 5An), which is another hallmark of aggresome formation (Johnston et al., 1998). The colocalization with ubiquitin conjugates (Fig. 5Aq-t) and 20S proteasome (Fig. 5Au-x) further substantiates the evidence that the observed inclusion bodies are indeed bona fide aggresomes.

To exclude the possibility that the localization of FAT10 to aggresomes is only an artifact of overexpression, we treated cells with TNF-α and IFN-γ to induce expression of FAT10, and examined whether the endogenous protein behaved similarly to the overexpressed version. The localization of both exogenous and cytokine-induced FAT10 in cells not treated with proteasome inhibitor was absolutely identical (compare Fig. 5Aa and 5Ba), and, although the aggresomes formed in the absence of overexpression tended to be smaller, there is no doubt that endogenously expressed FAT10 also localized to aggresomes (Fig. 5Be-h).

Since HEK293T cells produce extremely low levels of FAT10 conjugates (Hipp et al., 2005; Chiu et al., 2007), we assume the FAT10 observed in Fig. 5 to be mostly monomeric. To investigate whether FAT10-linked proteins are also transported to the aggresome, we transfected cells with our model conjugate – a linear, N-terminal fusion of FAT10 to GFP. This model conjugate is rapidly degraded by the proteasome (Hipp et al., 2005) and, as shown in Fig. 6Am-p, also localizes to the aggresome after proteasome inhibition. In fact, cells transfected with wild-type GFP did form aggresomes under proteasome inhibition, but these did not contain GFP (Fig. 6Ae-h), suggesting that FAT10-mediated targeting to the aggresome is specific and not an artifact of protein overexpression in the face of proteasome inhibition. Quantification of three independent experiments revealed a statistically significant increase in the formation of GFP-containing aggresomes only after proteasome inhibition and transfection with FAT10-GFP but not GFP alone (Fig. 6B). Taken together, these data provide compelling evidence for a specific localization of FAT10 to the aggresome after proteasome inhibition.

To test whether FAT10 also relies on the tubulin network for transport to aggresomes, we treated cells with the microtubule-depolymerizing agent nocodazole in addition to proteasome inhibitors and observed the effect that this had on the localization of FAT10. In accordance with the central role of retrograde microtubule-dependent transport in the formation of aggresomes (Kopito, 2000), depolymerization of microtubules also led to the disruption of FAT10-containing aggresomes. Instead of localizing to one prominent juxtanuclear aggresome upon proteasome inhibition (Fig. 7Ad-f), FAT10 accumulated under nocodazole treatment in several microaggregates, which were evenly dispersed throughout the cytoplasm (Fig. 7Ag-i).

Since catalytic activity of HDAC6 is required for the transport of polyubiquitylated proteins to the aggresome (Kawaguchi et al., 2003) but is dispensable for the interaction with FAT10 (Fig. 4; supplementary material Fig. S2), we set out to investigate the effect of the HDAC6-specific inhibitor tubacin (Haggarty et al., 2003) on the formation of FAT10-containing aggresomes. Unfortunately, though, we were not able to draw many conclusions regarding the formation of aggresomes from this experiment, since the combination of tubacin and MG132 led to profound cell death in FAT10-transfected cells only a few hours after treatment. Up until that point, however, the formation of aggresomes appeared to be comparable between cells treated with tubacin or the inactive control compound niltubacin (data not shown). Thus, it is clear that the transport of FAT10 to the aggresome requires a functional microtubule network, but the issue of whether it also requires the deacetylase function of HDAC6 remains unsolved.

To assay whether HDAC6 is essential for the localization of FAT10 to the aggresome, we set out to investigate the formation of aggresomes and the subcellular localization of FAT10 in cells derived from mice lacking the Hdac6 gene (Zhang et al., 2008). Cells stably expressing an siRNA directed against Hdac6 have previously been shown to be defective in the formation of polyubiquitin-containing aggresomes. After inhibition of the proteasome, in comparison to wild-type cells, fewer of these cells contained aggresomes and those aggresomes that did form were smaller in size (Kawaguchi et al., 2003). Our experiments revealed HDAC6-deficient cells to have a similar phenotype. When treated with 5 μM of the proteasome inhibitor MG132 for 4 hours, HDAC6 knock-out cells were still able to form polyubiquitin-containing aggresomes (Fig. 7B). However, fewer cells (approximately 40% less) contained an aggresome in comparison to HDAC6 wild-type cells (Fig. 8B) and the aggresomes that still formed were significantly smaller in size (Fig. 8C). In those cells that were transiently transfected with HA-FAT10, the subcellular localization – including the localization to aggresomes – of FAT10 mirrored that of polyubiquitin (Fig. 7Bh and 7Bp). The quantification of aggresome size as determined by anti-HA immunofluorescence also revealed HDAC6 knock-out cells to have significantly smaller FAT10-containing aggresomes (Fig. 8D). As cells that lack HDAC6 are partially deficient in the formation of FAT10-containing aggresomes – as well as polyubiquitin-containing aggresomes – HDAC6 certainly has an important role in their formation, even though it does not appear to be essential for the transport of neither polyubiquitin nor FAT10 to the aggresome or the formation of aggresomes in general.

Since the isolated ubiquitin-like domains of FAT10 exhibit differences in their ability to be degraded by the 26S proteasome and to interact with NUB1L (Schmidtke et al., 2006), we decided to examine whether there were also differences in their ability to bind to HDAC6 or to localize to aggresomes. To investigate the interaction with HDAC6, we performed co-immunoprecipitation experiments similar to the ones described above, only this time we used wild-type HDAC6 and N-terminal fusions of either full-length FAT10 or the isolated N- and C-terminal ubiquitin-like domains of FAT10 with GFP. We found that GFP-fusions with either the N- or C-terminal domain of FAT10, or full-length FAT10, were able to interact with HDAC6 under proteasome inhibition (Fig. 9A). In addition, immunofluorescence studies revealed that both domains localized to aggresomes under proteasome inhibition (Fig. 9B). We can therefore conclude that each ubiquitin-like domain of FAT10 is sufficient to interact with HDAC6 and mediate localization to the aggresome.

FAT10 is unique among the family of ubiquitin-like proteins because it is the only member – aside from ubiquitin – that can lead conjugated proteins directly to proteasomal degradation. Here, we report the discovery of a third non-covalent interaction partner of FAT10, the cytosolic deacetylase HDAC6. HDAC6 has a binding domain for ubiquitin-like proteins in its C-terminus that belongs to the family of `ubiquitin carboxyl-terminal hydrolase-like zinc finger' (Znf-UBP) domains, also known as either `deacetylase/ubiquitin-specific protease' (DAUP) domain, `polyubiquitin-associated zinc finger' (PAZ) domain, or `binder of ubiquitin zinc finger' (BUZ) domain.

Since NUB1L leads to an accelerated degradation of FAT10 (Hipp et al., 2004) and HDAC6 has been shown to be involved in the turnover of polyubiquitylated proteins (Kawaguchi et al., 2003; Boyault et al., 2006), we investigated whether the overexpression of HDAC6 would have an effect on the rate of FAT10 degradation – but this turned out not to be the case (Fig. 1C). We were able to obtain evidence that FAT10 is acetylated (supplementary material Fig. S3), which may in part account for the isoelectric variants observed for FAT10 on 2D gels (Raasi et al., 2001). However, the degree of acetylation was neither affected by the overexpression of HDAC6 nor by treatment of cells with the broad-range deacetylase inhibitor TSA, suggesting that the acetylation of FAT10 is not under dynamic regulation and that FAT10 is not a deacetylation substrate of HDAC6. Acetylation of FAT10 was also not required for the interaction with HDAC6 because (1) neither mutation of both active sites of HDAC6 nor treatment with TSA or tubacin abolished FAT10 binding (Fig. 4A-C), (2) the mutation of all lysine residues of FAT10 did not alter its association with HDAC6 (Fig. 4D), and (3) acetylated FAT10 did not preferentially associate with HDAC6 (Fig. 4E).

Much more revealing were our experiments investigating the hypothesis that HDAC6 might also shuttle FAT10 to the aggresome in situations of misfolded-protein stress or proteasome inhibition (Johnston et al., 1998; Kawaguchi et al., 2003). Indeed, proteasome inhibition led to the accumulation of FAT10 or FAT10-GFP in a single, large juxtanuclear inclusion body. Compelling evidence has been collected that this inclusion body is an aggresome: first, FAT10 colocalized in this structure with typical constituent proteins of aggresomes, such as HDAC6, γ-tubulin, polyubiquitin and the proteasome (Fig. 5). Second, FAT10 localized to this inclusion body only when the proteasome was inhibited, and did so with kinetics similar to the one reported for aggresome formation (Garcia-Mata et al., 1999) (Fig. 5 and data not shown). Third, the presence of an intact microtubule network, which is essential for the formation of aggresomes, was also required, because treatment of cells with the microtubule-depolymerizing agent nocodazole in addition to proteasome inhibition resulted in the formation of multiple FAT10-containing microaggregates in the cytosol (Fig. 7A). Evidence that the transport of FAT10 to aggresomes is not merely an artifact of overexpression comes from experiments in which endogenous FAT10, which was physiologically induced with IFN-γ and TNF-α, displayed a localization that was similar to that of the transiently transfected protein (Fig. 5B). In addition, co-immunoprecipitation also revealed endogenous, cytokine-induced FAT10 to be associated with HDAC6 (Fig. 1B).

The issue of whether transport of FAT10 to the aggresome is solely dependent on HDAC6 was addressed in cells derived from HDAC6-deficient mice. A previous study with cells stably expressing an shRNA directed against Hdac6 (Kawaguchi et al., 2003) revealed HDAC6 to be required for the proper formation of polyubiquitin-containing aggresomes, because HDAC6 knock-down cells contained fewer and smaller aggresomes as compared with wild-type cells. However, it was impossible to determine whether the residual formation of aggresomes was due to an incomplete knock-down of HDAC6 or a functional redundancy. Our experiments in cells lacking the Hdac6 gene now point towards the latter, as these cells display a similar phenotype (Fig. 7B; Fig. 8B,C). We were able to demonstrate the transport of FAT10 and polyubiquitin to the aggresome to be equally dependent on HDAC6; both FAT10- and polyubiquitin-containing aggresomes were significantly reduced in size in HDAC6-deficient compared with wild-type cells (Fig. 8C,D).

The role of tubulin deacetylation in the transport of FAT10 to the aggresome, however, remains unclear. By using the HDAC6-specific inhibitor tubacin, we attempted to investigate whether the deacetylase activity of HDAC6 was required for the transfer of FAT10 into aggresomes. Unfortunately, cells treated with proteasome inhibitor and tubacin at the same time did not survive long enough to investigate aggresome formation. We only can say for certain that the catalytic activity of HDAC6 is dispensable for its interaction with FAT10, as both inhibition as well as mutation of the active sites failed to abolish this interaction (Fig. 4; supplementary material Fig. S2). In fact, the association of HDAC6 with FAT10 was even increased after treatment of cells with the broad-spectrum deacetylase inhibitor TSA (Fig. 4B), which might be attributed to an increase in α-tubulin acetylation as well as non-specific effects of TSA on other targets, especially as treatment of cells with tubacin had a much less pronounced effect (Fig. 4C).

The question arises, why can the interaction between HDAC6 and FAT10 only be observed under proteasome inhibition? Boyault et al. have proposed a model in which HDAC6 senses an overload of polyubiquitylated proteins through the binding of polyubiquitin to its BUZ domain (Boyault et al., 2007). This then results in the release of heat-shock factor 1 and HDAC6 from a complex also containing p97 (VCP) and HSP90 (Boyault et al., 2007). An alternative model (Pai et al., 2007) suggests that it is not the accumulation of polyubiquitylated proteins but rather the lack of unconjugated ubiquitin resulting from proteasome inhibition that facilitates the interaction of HDAC6 and polyubiquitin. Owing to a much higher affinity of the BUZ domain for the C-terminus of ubiquitin rather than for polyubiquitin conjugates (Reyes-Turcu et al., 2006), the majority of HDAC6 would normally be bound to free ubiquitin. Under conditions of proteasome impairment, the decline in the level of monomeric ubiquitin releases HDAC6. In both models, HDAC6 would then be free to interact with polyubiquitin and also FAT10 conjugates.

It is interesting that FAT10, which can target proteins for proteasomal degradation independently of the ubiquitin system, uses the same rescue strategy when the proteasome is overwhelmed. This implies that the rapid removal of FAT10 and/or its conjugates is essential for cell survival – for example, to protect it from the deleterious effects caused by an excess of free FAT10. This is suggested by the finding that the ectopic expression of FAT10 induces apoptosis in several cell types (Raasi et al., 2001; Ross et al., 2006).

The formation of ubiquitin-containing aggregates is a hallmark of many neurodegenerative diseases, such as Huntington disease, PD and DLB. It is notable that NUB1L, another non-covalent interaction partner of FAT10 (Hipp et al., 2004), accumulates in Lewy bodies in PD and DLB patients (Tanji et al., 2007) and localizes to aggresomes under proteasome inhibition (data not shown). It will hence be interesting to investigate whether FAT10 is also a component of Lewy bodies and could even be responsible for targeting NUB1L to these sites. Since the basal expression of FAT10 in the brain is extremely low (Canaan et al., 2006), an involvement of FAT10 would rely on a strong induction of Fat10 expression with TNF-α and IFN-γ, which is probably not prevalent in these chronic neurological disorders. Recently, FAT10 was shown to be highly upregulated in hepatocellular carcinoma (Lee et al., 2003) and, in a drug-induced mouse model of hepatocellular carcinoma, FAT10 localized to Mallory Denk bodies, which are a special form of aggresomes observed in a variety of liver diseases (Oliva et al., 2008; Zatloukal et al., 2006). By contrast, the localization of the Fat10 gene in the major histocompatibility class I locus and the fact that it is inducible by proinflammatory cytokines point towards a function in the immune response. It remains subject to investigation whether FAT10 is perhaps involved in the formation of DALIS (dendritic cell aggresome-like induced structures) – transient aggresome-like structures that function as storage compartments of antigenic proteins during the maturation of dendritic cells (Lelouard et al., 2002). In addition, because autophagy has been implicated in the innate as well as the adaptive immune response to intracellular pathogens, it is tempting to speculate that FAT10 might be involved in those processes as well.

Spontaneously immortalized Hdac6^+/+ and Hdac6^–/– mouse embryonic fibroblasts (MEFs) were generously provided by Patrick Matthias (Zhang et al., 2008). MEFs and HEK293T cells were cultivated in DMEM supplemented with 10% FCS and 100 μg/ml penicillin/streptomycin (all from Invitrogen). HEK293T cells were transiently transfected using FuGENE 6 (Roche) according to the manufacturer's instructions. MEFs were transiently transfected using the Amaxa MEF2 kit according to the manufacturer's instructions. For the induction of endogenous FAT10, cells were treated for 16-20 hours with 500 U/ml human TNF-α (Endogen) and 200 U/ml human IFN-γ (Endogen).

The yeast two-hybrid screen was performed as previously described (Hipp et al., 2004).

The antibodies used in this study were anti-FLAG M2 (Sigma), anti-FLAG polyclonal (Sigma), anti-HA HA-7 (Sigma), anti-HA polyclonal (Sigma), anti-AcK 103 (Cell Signaling), anti-γ-tubulin GTU-88 (Sigma), anti-ubiquitin FK2 (Biomol), anti-HC9/20S (Affinity), anti-HDAC6 polyclonal (Cell Signaling) and anti-GFP polyclonal (Sigma). The anti-FAT10 polyclonal antibody was generated as previously described (Hipp et al., 2005). Horseradish-peroxidase-coupled anti-HA HA-7 antibody was purchased from Sigma, and anti-rabbit and anti-mouse antibodies from DAKO. The antibodies used in immunofluorescence – anti-HA 16B12 (Alexa-Fluor-488-coupled), goat anti-rabbit IgG (H+L) (Alexa-Fluor-488-coupled) and goat anti-mouse IgG (H+L) (Alexa-Fluor-546-coupled) – were purchased from Molecular Probes. MG132, nocodazole and TSA were purchased from Sigma.

pcDNA3-FLAG-tagged wild type (wt), ΔBUZ, 1-840, 1-503 and ΔN HDAC6, as well as the catalytically dead mutant, were a kind gift from Tso-Pang Yao (Kawaguchi et al., 2003). The additional mutants were constructed by first inserting a FLAG-tag (5′-ATGGACTACAAGGACGACGATGACAAG-3′) between the KpnI and BamHI restriction sites of pcDNA3.1. The desired fragments of HDAC6 were then amplified by PCR using pcDNA3/FLAG-HDAC6 wt as a template and cloned into the vector via BamHI and XbaI restriction sites. The following primers were used for PCR: forward 5′-TAACGGATCCACCTCAACCGGCCAGGATTC-3′, reverse 5′-TAAGTCTAGATTAGAAGCTGTCATCCCAGAGGCA-3′ (1-105); forward 5′-TAACGGATCCTCTCCCAGTACACTGATTGGG-3′, reverse 5′-TAAGTCTAGATTAGTGTGGGTGGGGCATATCCTC-3′ (BUZ); forward 5′-GTATGGATCCGACAGAGAAGGACCCTCCAG-3′ (841-1215); forward 5′-TAACGGATCCCCGGAAGGCCCTGAGCGG-3′, reverse 5′-TAACTCTAGATTAAAGAACCTCCCAGAAGGGCTCA-3′ (CAT1); forward 5′-TCTAAGATCTGTACCCCAGCGCATCTTGC-3′, reverse 5′-ACAGTCTAGATTATTCTACCTTCATGACCCGTAAG-3′ (CAT2). The catalytically inactive CAT1 domain was generated using the catalytically dead full-length protein as a template. All sequences were verified by dideoxy sequencing. pcDNA3.1/HA-FAT10 was generated as previously described (Raasi et al., 2001), as were pcDNA3.1/HA-FAT10 K0 (lysine-less) (Hipp et al., 2005), pEGFP-N1/FAT10-AV-GFP and pGEX-4T-3/GST-FAT10 (Hipp et al., 2004), as well as EGFP-N1/FAT10-N-GFP and pEGFP-N1/FAT10-C-GFP (Schmidtke et al., 2006). pGEX-4T-3/GST-His₆-FAT10ΔGG was generated by mutating the C-terminal diglycine to AV. pEGFP-N1 was purchased from Clontech.

Pulse-chase experiments were performed as previously described (Hipp et al., 2004). For the immunoprecipitation experiments, transiently transfected cells were incubated for 6 hours in the presence of 5 μM MG132, 1 μM nocodazole, 5 μM TSA or the respective solvent as a negative control. After harvesting, cells were lysed in 20 mM Tris/HCl pH 7.8 containing 0.1% Triton X-100, 1 μM pepstatin, 10 μM leupeptin, 5 μg/ml aprotinin, 100 μM PMSF, 20 μM MG132, 2 mM ATP and 2 mM MgCl₂ for 30 minutes on ice, followed by sonication, addition of 150 mM NaCl and centrifugation for 15 minutes at 20,000 g. Cleared lysates were subjected to pre-clearance with either protein A or protein G affinity gel (Sigma) for 30 minutes at 4°C followed by incubation over night at 4°C with 10 μg anti-FLAG M2 antibody and protein G or 15 μl anti-FAT10 or control serum and protein A, respectively. Immunoprecipitates were washed five times in lysis buffer and analyzed by SDS-PAGE and western blotting. For the experiments examining the acetylation of FAT10, 5 μM TSA was included in the lysis buffer.

GST-FAT10 and GST were purified as previously described (Hipp et al., 2005). FLAG-tagged wild-type as well as mutant HDAC6 was in vitro transcribed and translated using the TNT T7 Coupled Reticulocyte Lysate System (Promega) according to the manufacturer's instructions. After pre-clearance with GSH-Sepharose (GE Healthcare), half of each reaction was incubated with 30 μl GSH-Sepharose (blocked with 1 mg/ml BSA) and either 50 μg GST-FAT10, GST-His₆-FAT10 (ΔGG) or GST as a negative control in 200 μl of the lysis buffer used in the immunoprecipitation experiments. Samples were incubated over night at 4°C, washed five times and analyzed by SDS-PAGE followed by autoradiography.

For the anti-γ-tubulin (dilution 1:500) and anti-20S proteasome (dilution 1:100) stainings, HEK293T cells were grown on poly-L-lysine (Fluka)-coated coverslips, incubated with 5 μM MG132 or solvent only for 6 hours and fixed with ice-cold MeOH:acetic acid (3:1) for 10 minutes. For the anti-FLAG M2 (dilution 1:300), anti-ubiquitin FK2 (dilution 1:100), anti-GFP (dilution 1:200) and anti-FAT10 (dilution 1:200) stainings, cells were grown on poly-L-lysine-coated coverslips, incubated with 5 μM MG132, 1 μM nocodazole or solvent only for 6 hours (HEK293T cells) or 4 hours (MEFs), fixed with 4% paraformaldehyde for 30 minutes and permeabilized with 0.1% Triton X-100 for 10 minutes at room temperature. Cells were first labeled with the primary antibodies, followed by washing and incubation with the respective Alexa-Fluor-labeled secondary antibodies (dilution 1:300), followed by washing and incubation with the Alexa-Fluor-labeled anti-HA antibody (dilution 1:300) where applicable. All antibodies were diluted in 0.2% gelatine, except for the anti-γ-tubulin and anti-FAT10 stainings, for which 1.5% normal goat serum was added. All incubations were carried out for 1 hour at room temperature. Images were acquired and analyzed with an LSM 510 confocal laser-scanning microscope (Carl Zeiss) using a 63× plan-apochromat oil-immersion objective (NA=1.4). Statistical analysis of the number of aggresome-containing cells was performed by randomly selecting approximately 100 cells and scoring them for the presence of a large, juxtanuclear aggresome. Statistical analysis of aggresome size was carried out by randomly selecting approximately 50 (for the anti-ubiquitin staining) or ten (for the anti-HA staining after HA-FAT10 transfection) aggresome-containing cells. Aggresome sizes were then quantified using ImageJ (http://rsb.info.nih.gov/ij/).

We thank Patrick Matthias for the donation of Hdac6^–/– cells, Tso-Pang Yao for the donation of plasmids, Matthias Langhorst for the kind instruction in confocal microscopy, and Stuart L. Schreiber and Ralf Mazitschek for the gift of tubacin and niltubacin. This work was supported by the German Research Foundation (DFG) grant GR 1517/2-3.