Mimicking proteasomal release of polyglutamine peptides initiates aggregation and toxicity

Numerous neurodegenerative diseases are manifested by the accumulation and aggregation of intracellular proteins. These diseases include polyglutamine (polyQ) expansion disorders such as Huntington's disease (HD), spinal bulbar muscular atrophy (SBMA) and various spinocerebellar ataxias (SCAs). PolyQ disorders are dominantly inherited and caused by expansions of CAG repeats. Normally, the disease-related proteins involved contain sequences of 6-40 glutamine repeats, but expansion of these tracts to 40-300 repeats leads to disease. The age of onset of the disorder is inversely correlated with the repeat length of the polyQ tracts (reviewed by Orr and Zoghbi, 2007).

The presence of proteolytic protein fragments harbouring a polyQ tract in aggregates (DiFiglia et al., 1997; Goti et al., 2004; Li et al., 1998; Schmidt et al., 1998) has led to the `toxic fragment hypothesis', which states that proteolytic fragments of polyQ-expanded huntingtin (Cooper et al., 1998), androgen receptor (Merry et al., 1998) or certain ataxins (Haacke et al., 2006; Ikeda et al., 1996; Young et al., 2007) initiate protein aggregation and induce neuronal toxicity. Full-length polyQ proteins aggregate, but at a much slower rate than their proteolytic fragments (Merry et al., 1998). These fragments can be generated by proteases such as caspases, aspartic endopeptidases, calpains and the proteasome (Gafni et al., 2004; Goldberg et al., 1996; Graham et al., 2006; Lunkes et al., 2002; Wellington et al., 1998). Accumulation of these proteolytic fragments may therefore function as a nucleation centre that sequesters full-length polyQ proteins in time. The proteasome can degrade both wild-type and expanded forms of most polyQ proteins, as was demonstrated in cultured cells and animal models (Bence et al., 2001; Jana et al., 2005). Surprisingly, polyQ-expanded proteins are not degraded to completion by the proteasome either in vitro and in vivo (Holmberg et al., 2004; Venkatraman et al., 2004). Venkatraman and colleagues showed that isolated proteasomes cannot digest polyQ tracts present within a protein, which will result in the release of polyQ peptides. Although flanking amino acids may be removed by exo-peptidases, the polyQ tracts themselves will accumulate when not efficiently cleared by downstream peptidases.

To examine the fate of these polyQ peptides downstream the proteasome, we mimicked proteasomal generation of polyQ peptides in living cells. If polyQ peptides are degradation-resistant upon release into the cytoplasm, they may subsequently accumulate and initiate aggregation. Most studies investigating polyQ disorders use polyQ constructs that contain a starting methionine and/or fusion tags such as fluorescent proteins. These polyQ constructs, including polyQ-GFP, huntingtin exon-1 or their short-lived variants, will require processing by the proteasome and therefore do not represent polyQ peptides generated by the proteasome. To mimic pure polyQ peptide generation, we generated a fusion protein containing green fluorescent protein, ubiquitin and polyQ peptides (GFP-Ub-polyQ). This fusion protein efficiently releases non-tagged polyQ peptides upon cleavage by ubiquitin C-terminal hydrolases (Johnson et al., 1995). We show that upon release, only polyQ peptides of disease-related lengths accumulated inside cells, and initiated intracellular protein aggregation. Proteasomes were rapidly sequestered, followed by ubiquitylated proteins, and associated chaperones, as has been observed in various polyQ disorders (Haacke et al., 2006; Holmberg et al., 2004; Kim et al., 2002). Also, various proteins containing either wild-type or expanded polyQ stretches were sequestered (Haacke et al., 2006; Perez et al., 1998). In addition, accumulation of expanded polyQ peptides led to neuronal toxicity.

To examine the fate of proteasomal-released polyQ peptides in living cells, we generated fusion proteins of fluorescently tagged Ub with polyQ peptides of wild-type and disease-related lengths. Upon expression, the C-terminal polyQ peptide will be efficiently released from GFP-Ub by immediate cleavage by ubiquitin C-terminal hydrolases (Johnson et al., 1995). As a result, the generated polyQ peptide does not contain a starting methionine residue, which may affect degradation properties because of similarities with the N-terminus of a full-length protein (Bachmair et al., 1986). Also, no tags (such as fluorophores or antibody epitopes) were directly attached to the polyQ peptide. As Ub was fluorescently tagged, the fluorescence intensity reflected the amount of generated polyQ peptides. PolyQ peptides of 16, 65 or 112 glutamine residues were fused to GFP-Ub resulting in GFP-Ub-Q16, GFP-Ub-Q65 and GFP-Ub-Q112, respectively (Fig. 1A).

Expression of the different GFP-Ub-polyQ proteins and subsequent release of polyQ peptides were analyzed 48 hours after transfection. Western blot analysis demonstrated the presence of GFP-Ub (36 kDa) separated from all polyQ proteins (Fig. 1B, left panel). In addition, a large Ub conjugate was present, as shown before for GFP-Ub (Dantuma et al., 2006). No additional bands were detected that could represent uncleaved GFP-Ub-polyQ proteins. Efficient cleavage was also observed when the western blot was analysed for Ub (Fig. 1B, right panel). These results indicate that all polyQ peptides were efficiently cleaved from the GFP-Ub protein. Subsequent immunoblotting against polyQ using the antibody 1C2 (Trottier et al., 1995) showed that polyQ peptides were present in the GFP-Ub-Q65 and GFP-Ub-Q112 lanes (Fig. 1C). The mobility on SDS-PAGE of expanded polyQ peptides was different from their calculated molecular masses, as has been observed before for polyQ-containing proteins (Holmberg et al., 2004; Servadio et al., 1995). Some additional high molecular mass bands were present, which may represent oligomeric polyQ structures, as these bands were GFP and Ub negative (Fig. 1B). The absence of Q16 peptides in cells expressing GFP-Ub-Q16 indicates that small Q peptides are efficiently cleared from the cytoplasm. It is unlikely that small Q peptides are not recognized by the 1C2 antibody, since a Q16-GFP fusion protein was recognized by 1C2 with almost equal efficiency as expanded GFP-polyQ fusions (supplementary material Fig. S1A). Accumulation of Q65 and Q112 peptides, but not of Q16 peptides, suggests that expanded polyQ peptides were not efficiently degraded in living cells. To our knowledge, these peptides are the first group of peptides to be found to be resistant to degradation.

Since proteolytic protein fragments containing polyQ tracts are more prone to aggregation than the full-length protein, we examined whether the accumulation of Q65 and Q112 peptides initiated aggregate formation. We observed a similar intracellular distribution of GFP-Ub in cells transfected with either GFP-Ub or GFP-Ub-Q16. GFP-Ub was enriched in the nucleus but was also present in the cytoplasm and on vesicles (Fig. 1D), similar to the distribution of endogenous Ub (Dantuma et al., 2006; Qian et al., 2002). By contrast, expression of GFP-Ub-Q65 and GFP-Ub-Q112 resulted in the appearance in either the nucleus or cytoplasm of a distinct intracellular structure decorated with fluorescent Ub in a high percentage of the transfected cells. The number of cells containing these structures increased both in time and with polyQ length (Fig. 1E). To see whether the length-dependency of aggregate formation would also hold true for polyQ lengths near the threshold, we expressed GFP-Ub fused to polyQ peptides of 33 or 48 glutamine residues. Whereas GFP-Ub-Q33-expressing cells showed no aggregates, cells expressing GFP-Ub-Q48 had aggregates, although in a much lower percentage of cells than those expressing Q65 or Q112 peptides (supplementary material Fig. S1B; and data not shown). GFP-Ub fluorescence was usually present in a ring around a dark core, indicating that Ub was recruited (Fig. 1F). At the ultrastructural level, this structure was found to have a radiating dense core similar to aggregates formed by non-cleavable GFP-polyQ fusion proteins (Fig. 1F) and expanded huntingtin (Qin et al., 2004). In cells expressing Q65 and Q112 peptides, these dense structures were resistant to SDS and selectively trapped in a filter retardation assay (Wanker et al., 1999). Immunostaining using 1C2 showed that the trapped structures contained polyQ peptides (Fig. 1G), similar to huntingtin exon-1 Q103 (httex1-Q103-GFP) (Wanker et al., 1999). This suggests that expanded polyQ peptides induce intracellular SDS-resistant aggregates. Although Httex1-Q103-GFP was also positive for GFP, no GFP was present on the filter trap with the GFP-Ub-polyQ constructs, indicating efficient cleavage of the GFP-Ub-polyQ fusion proteins (Fig. 1G). Also, analysis of the soluble and insoluble fraction of cell lysates showed no uncleaved GFP-Ub-Q112 fusion proteins in either fraction (supplementary material Fig. S1C).

To confirm the presence of polyQ peptides in intracellular aggregates, we immunostained cells expressing Q16, Q65 or Q112 peptides with 1C2. As expected, no polyQ peptides were detected in cells expressing GFP-Ub or GFP-Ub-Q16 (Fig. 1H). However, cells transfected with GFP-Ub-Q65 or GFP-Ub-Q112 showed two patterns of polyQ staining, dependent on the presence of aggregates. When aggregates were not present, polyQ staining was mainly cytoplasmic, whereas GFP-Ub localization was predominantly nuclear. By contrast, cells containing polyQ peptide aggregates were not recognized by 1C2 (Fig. 1H, arrows indicate an aggregate). A similar difference in immunostaining was obtained using the anti-polyQ antibody MW1 (Ko et al., 2001) (supplementary material Fig. S1D). The absence of polyQ staining in cells containing aggregates is probably due to the dense aggregate structure and its surrounding protein layers that may shield the polyQ core. Indeed, pretreatment with proteinase K degraded shielding proteins and resulted in positive immunostaining of polyQ peptide aggregates (Fig. 1I), as has been observed previously for huntingtin aggregates (Qin et al., 2004). To further confirm that the aggregates contain polyQ peptides, we used a cyan fluorescent protein (CFP)-tagged Q-binding peptide (QBP1), which selectively binds to polyQ aggregates (Nagai et al., 2000). QBP1 showed a cytoplasmic distribution pattern when expressed alone or together with RFP-Ub or RFP-Ub-Q16 (data not shown). However, cells harbouring aggregates initiated by Q112 peptides showed binding of QBP1 to aggregates (Fig. 1J). Taken together, these results indicate that expanded polyQ peptides are not efficiently degraded and subsequently initiate formation of aggregates that display all the characteristics of disease-related polyQ aggregates.

Aggregates formed by expanded polyQ proteins often sequester proteins involved in the ubiquitin proteasome system (UPS) and also chaperones (Holmberg et al., 2004; Kim et al., 2002). We examined whether aggregates induced by expanded polyQ peptides showed a similar sequestration of UPS components. GFP-Ub was present in a ring around the aggregates (Fig. 1F). Absence of Ub in the aggregate core can be explained by the lack of lysine residues in polyQ peptides, thereby excluding ubiquitylation of the polyQ peptides. The presence of GFP-Ub around the core was not due to inefficient cleavage of GFP-Ub-polyQ, since no uncleaved GFP-Ub-polyQ fusions could be detected by SDS-PAGE (Fig. 1B; supplementary material Fig. S1C) and filtertrap (Fig. 1G). In addition, coexpression of GFP-Ub with RFP-Ub-Q112 showed a similar sequestration of both fluorescently tagged Ub proteins into aggregates (Fig. 2A), indicating efficient cleavage. This suggests that the presence of GFP-Ub is due to ubiquitylation of sequestered proteins.

We examined whether proteasomes colocalized with polyQ aggregates in our model, by coexpressing the different RFP-Ub-polyQ constructs with GFP-tagged immuno-proteasomal subunit LMP2. LMP2 is efficiently incorporated into active proteasomes (Reits et al., 1997). Notably, LMP2-GFP was present in the core of polyQ aggregates, suggesting that proteasomes were recruited to aggregates even before Ub sequestration (Fig. 2B). A similar recruitment was observed when using the constitutive proteasome subunit β7 (Fig. 2D). This finding most probably reflects a proteasomal attempt to degrade accumulating polyQ peptides. The sequestered proteasomes and Ub seemed irreversibly trapped, which was revealed when fluorescence recovery after photobleaching (FRAP) (Reits and Neefjes, 2001) was applied to determine on/off rates of the sequestered molecules. Upon photobleaching of one half of an aggregate, no exchange between the sequestered proteasomes or Ub and the surroundings was observed (supplementary material Fig. S2A). This indicates that the proteasome becomes immobilized, as has been previously observed (Holmberg et al., 2004).

We also examined whether chaperones such as Hsp70 were interacting with polyQ aggregates, as has been observed in polyQ diseases (Kim et al., 2002; Matsumoto et al., 2006). Upon co-transfection of the different RFP-Ub-polyQ fusion proteins with GFP-tagged Hsp70, we observed an additional ring-like structure of Hsp70-GFP around the Ub-positive aggregate (Fig. 2C). FRAP analysis revealed that Hsp70 was not irreversibly trapped in the aggregate (supplementary material Fig. S2A), consistent with previous observations (Kim et al., 2002; Matsumoto et al., 2006). To compare the composition of aggregates initiated by Q112 peptides with aggregates formed by polyQ-expanded huntingtin exon-1, cells were transfected with either GFP-Ub-Q112 or httex1-Q103-GFP together with β7-RFP-tagged proteasomes, and cells were subsequently immunostained for endogenous Hsp70. Triple colour analysis showed that the core of the aggregate was positive for proteasomes (red). This core was surrounded by Ub or httex1-Q103 (green). Finally, Hsp70 was present within the outermost layer of the aggregate (blue; Fig. 2D). This suggests that various proteins associate at different stages or with different affinities during aggregate formation. The presence of GFP-Ub and httex1-Q103 in a similar layer may suggest the recruitment of ubiquitylated proteins and polyQ proteins in this stage of aggregate formation. Since aggregates initiated by expanded polyQ peptides also contained Ub, proteasomes and chaperones as has been described before, our model faithfully mimics aggregate formation in polyQ diseases.

The presence of httex1-Q103 in ring-like structures around the aggregate and not within the core (Fig. 2D) suggests recruitment of large polyQ fragments into aggregates in a later stage. To examine this hypothesis, we coexpressed RFP-Ub-Q112 and httex1-Q103-GFP. Indeed, we found that httex1-Q103-GFP was sequestered into aggregates induced by polyQ peptides (Fig. 3A). In addition, the aggregation rate of httex1-Q103-GFP was also dramatically increased when Q112 peptides were present (supplementary material Fig. S2B), which suggests that polyQ peptides initiate aggregates that accelerate huntingtin aggregation. Similar results were obtained with truncated polyQ-expanded ataxin-3 (Atx3-Q85-GFP) and the SBMA-related truncated androgen receptor with a Q84 repeat (AR-Q84-GFP) (data not shown).

Aggregates induced by disease-related polyQ proteins also sequester the wild-type protein expressed by the non-expanded allele (Busch et al., 2003; Haacke et al., 2006). We examined whether polyQ peptide aggregates also sequester non-expanded, wild-type polyQ proteins. The non-expanded httex1-Q25-GFP remained freely distributed in cells that coexpressed either RFP-Ub or RFP-Ub-Q16 (supplementary material Fig. S2B). By contrast, httex1-Q25-GFP was recruited into polyQ peptide aggregates when co-transfected with RFP-Ub-Q112 (Fig. 3B; supplementary material Fig. S2B). A similar entrapment of wild-type truncated ataxin 3 (Atx3-Q28-GFP) (supplementary material Fig. S2C) and the truncated androgen receptor (AR-Q19-GFP) was observed (data not shown). This sequestration of wild-type polyQ proteins may therefore lead to loss of function. Sequestering of non-expanded polyQ proteins was not limited to disease-related proteins, as other polyQ proteins were recruited into aggregates initiated by polyQ peptides, including Q16-GFP (Fig. 3C), and also the Q-stretch-containing transcription factor TBP1 when nuclear aggregates were present (Fig. 3D).

To examine whether polyQ peptides also initiate aggregate formation in neuronal cells, we transiently transfected N2A neuroblastoma cells with the various GFP-Ub-polyQ constructs. N2A cells transfected with either GFP-Ub-Q65 or GFP-Ub-Q112 developed aggregates similar to those present in non-neuronal cells (Fig. 4A), whereas GFP-Ub-Q16-expressing cells showed a Ub distribution comparable to GFP-Ub alone. Since HD mostly affects striatal cells, we also used immortalized SThdh^+/+ striatal cells (Trettel et al., 2000), which similarly generated intracellular aggregates when transfected with GFP-Ub-Q65 or Q112 (Fig. 4A). Many cells rounded up after expression of expanded polyQ peptides, suggesting toxicity, although this did not correlate with the presence of GFP-Ub-positive aggregates. To determine whether the expressed polyQ peptides were toxic, the viability of transfected N2A cells was tested using propidium iodide (PI). Expression of expanded polyQ peptides resulted in increased numbers of PI-positive cells (data not shown). However, hardly any double-positive cells were observed. This is presumably explained by the fact that uptake of PI into polyQ peptide-expressing cells was often preceded by loss of GFP fluorescence (Fig. 4B) as observed previously (Arrasate and Finkbeiner, 2005). Because loss of fluorescence seemed to be associated with cell death, we used another approach to quantify polyQ peptide-induced toxicity. To determine changes in the number of GFP-Ub-positive cells with time, we used FACS analysis and compared cell populations expressing the different GFP-Ub-polyQ proteins at 24 and 48 hours after transfection. There was no difference in GFP-Ub fluorescence between cells expressing GFP-Ub or GFP-Q16 with time. However, a significant decrease in fluorescence was observed in cells expressing GFP-Ub-Q112 when compared with GFP-Ub or GFP-Ub-Q16 (P<0.05), indicating that expression of Q112 peptides induced cell death (Fig. 4C). GFP-Ub-Q65 had a mild, although not significant, effect on cell death. Taken together, these results showed that expanded polyQ peptides form aggregates and become toxic to neuronal cells.

Proteolytic fragments containing expanded polyQ tracts are more aggregation-prone than original full-length proteins, as has been shown for huntingtin (Cooper et al., 1998), androgen receptor (Merry et al., 1998), ataxin 3 (Haacke et al., 2006) and ataxin 7 (Young et al., 2007). Recently, it was also postulated that an expanded polyQ fragment was expressed in SCA8, because antisense transcription resulted in polyQ inclusions (Ikeda et al., 2007). These data suggest that polyQ fragments may be fundamental in initiating aggregation. It has, however, been shown that expanded polyQ proteins are efficiently targeted to the proteasome (Holmberg et al., 2004), which can degrade entire proteins, with the exception of polyQ tracts (Venkatraman et al., 2004). Degradation by the proteasome may result in the release of polyQ peptides, whose flanking amino acids may be removed by exo-peptidases. It is unknown whether the resulting pure polyQ peptides are rapidly degraded by peptidases. If resistant, their subsequent accumulation may initiate aggregation and toxicity as observed in polyQ disorders. In order to examine this toxic fragment hypothesis, we mimicked intracellular proteasomal polyQ peptide generation as closely as possible by fusing pure polyQ peptides to GFP-tagged Ub. Although Ub-polyQ fusions have been used before, these polyQ fragments also included either GFP tags (Kaytor et al., 2004; Verhoef et al., 2002) or additional amino acids including a starting methionine residue (Marsh et al., 2000). Expression of our constructs resulted in the efficient release of `naked' polyQ peptides because of immediate cleavage by Ub C-terminal hydrolases. This was shown by SDS-PAGE, in which a GFP-Ub band was present at the same position irrespective of the original construct (Fig. 1B), by a filter retardation assay (Fig. 1G) and examination of the insoluble fraction (supplementary material Fig. S1C), and by different intracellular locations of GFP-Ub and polyQ peptides (Fig. 1H). Since the released polyQ peptides do not contain a starting methionine or additional tags, they closely resemble peptide generated by the proteasome. All previous studies have relied on expression of polyQ fusions that did include such features, which can significantly alter the in vivo behaviour of polyQ fragments. Starting methionines will make the peptides resemble the N-terminus of proteins, possibly affecting the rate of degradation (Bachmair et al., 1986). Fluorescent tags contain lysine residues, which can serve as targets for ubiquitylation and subsequent degradation by the proteasome. The intracellular release of monomeric polyQ peptides is also closer to the in vivo situation than the addition of synthesized polyQ peptide aggregates to cells (Yang et al., 2002).

We showed that only polyQ peptides with Q repeat lengths similar to disease-related peptides accumulated in the cell and initiated aggregation. The characteristics of aggregates induced by expanded polyQ peptides were similar to those of aggregates initiated by expression of expanded polyQ-containing proteins (Holmberg et al., 2004; Kim et al., 2002; Perez et al., 1998; Qin et al., 2004). These characteristics include sequestration of proteasomes, ubiquitin and other polyQ-containing proteins such as TBP, and the presence of Hsp70. Although previous studies only speculated on the effect of proteasomal release of polyQ peptides in living cells, we show here that `proteasomal-derived' expanded polyQ peptides by themselves are sufficient to accumulate and initiate aggregation. Accumulation of expanded polyQ peptides is toxic to neuronal cells, but it remains to be established which particular step in aggregate formation is toxic. The toxicity seems to be induced by necrosis instead of apoptosis, as no apoptotic markers such as annexin 5 or activated caspases were detected (data not shown). The toxic species may be either small polyQ peptide oligomers or large polyQ aggregates. Further studies are required to determine whether the proteasome can indeed generate similar polyQ peptides from different polyQ proteins. If so, these released polyQ peptides may be the common feature of the different polyQ disorders.

Based on our findings, we propose a model in which expanded polyQ peptides are resistant to degradation, and their accumulation leads to intracellular polyQ aggregates (Fig. 5). Proteasomes are rapidly recruited into the polyQ core, possibly in a final attempt to degrade the expanded polyQ peptides. Subsequently, other proteins are sequestered and ubiquitylated, perhaps because of (partial) unfolding. These events also lead to the binding of chaperones such as Hsp70 that may recognize denatured proteins. All these events result in concentric ring-like structures formed around the aggregate (Fig. 5). Essential proteins are depleted from the cell, contributing to cellular dysfunction. We conclude that polyQ peptides may be fundamental in initiating aggregation and sequestration of different types of proteins including polyQ proteins. Although FRAP experiments indicated that UPS components were immobilized, we could not detect proteasomes or Ub in the SDS-insoluble fraction of cell lysates (data not shown) or on filter traps (Fig. 1G). This suggests that the recruited UPS components can still be solubilized.

We were able to detect expanded polyQ peptides containing Q65 or Q112 on western blots and by immunostaining in fixed cells, but we were unable to detect any Q16 peptides. These short polyQ peptides are most probably rapidly degraded by downstream peptidases such as PSA (Bhutani et al., 2007) that can digest short polyQ peptides and perhaps also extended peptides, with less efficiency. Alternatively, a technical explanation for this result might be poor staining by 1C2. It has been suggested that anti-polyQ antibodies do not detect the polyQ peptide itself, but interact with the secondary structure created by the expanded polyQ peptide (Li et al., 2007). Nonetheless, we showed that the 1C2 antibody was able to recognize a Q16 peptide fused to GFP with almost equal efficiency as GFP-Q65 and GFP-Q112 proteins. Similarly, the polyQ antibody MW1 was able to detect a Q16-GFP fusion protein but no Q16 peptides derived from GFP-Ub-Q16 (data not shown). This shows our inability to detect Q16 peptides is not likely to be caused by the intrinsic inability of 1C2 to recognize this peptide species. Thus, the inability to detect any Q16 peptides in cells expressing GFP-Ub-Q16 is most probably due to rapid and efficient degradation of non-expanded polyQ peptides. During the preparation of this article, work has been published that suggests that isolated proteasomes are able to cleave multiple times within a short polyQ-containing peptide (Pratt and Rechsteiner, 2008). They argued that Venkatraman and collegues (Venkatraman et al., 2004) underestimated the amount of cleaved polyQ fragments as a consequence of their mass-spectrometry methods. However, their conclusion was also based on other experiments such as western blot analysis of polyQ protein products generated by proteasomes, and are in line with the conclusions by Holmberg and colleagues (Holmberg et al., 2004). The experiments of Pratt and Rechsteiner (Pratt and Rechsteiner, 2008) were done in the presence of a mutated PA28γ subunit, which alters proteasomal access and specificity to peptides. In addition, although isolated proteasomes may be able to cleave short polyQ peptides, our observation that Q65 and Q112 peptides readily aggregate suggests that the proteasome cannot efficiently degrade expanded polyQ peptides and thus cannot prevent their accumulation.

PolyQ aggregation is commonly visualized using full-length or truncated polyQ proteins that are GFP tagged, which therefore represent proteins that require degradation by the proteasome. When such GFP-tagged proteins are degraded by the proteasome, this results in the release of non-fluorescent polyQ peptides that will initiate aggregation and subsequently sequester GFP-tagged fragments or full-length polyQ proteins. Visualization of aggregation using GFP-tagged polyQ proteins thus represents a later stage in aggregate formation and does not reveal much about the initiation of aggregation. Long- or short-lived polyQ proteins have been used to link degradation to aggregation kinetics, where long-lived GFP-polyQ (Michalik and Van Broeckhoven, 2004; Verhoef et al., 2002) and GFP-tagged polyQ-expanded htt fragment (Kaytor et al., 2004) fusion proteins were compared to short-lived variants. In these studies proteasomal degradation of short-lived expanded polyQ proteins resulted in reduced formation of GFP-positive aggregates compared with their long-lived counterparts. Strikingly, toxicity was higher in cells expressing the short-lived expanded htt fragment than in cells expressing the long-lived version (Kaytor et al., 2004). Our model can explain this unexpected finding: short-lived polyQ proteins are more rapidly degraded than long-lived proteins, resulting in aggregation-prone and toxic polyQ peptides. However, such aggregates remain invisible as GFP fluorescence of the short-lived proteins is lost by rapid breakdown, preventing its incorporation in the aggregates. Consequently, this has probably led to an underestimation of the real number of aggregates formed by short-lived proteins in these studies. The increased toxicity was in fact presumably caused by higher levels of generated polyQ peptides. The reduced toxicity in GFP-positive cells found by Verhoef and colleagues (Verhoef et al., 2002) may similarly be explained by preferential loss of fluorescence by toxic fragments, since only toxicity of GFP-positive cells were analyzed.

Our method mimicking proteasomal release of polyQ peptides is also a valuable tool to investigate a number of important questions concerning the role of polyQ peptides in HD and related neurodegenerative disorders. It enables us to identify proteases or peptidases that can target intracellular polyQ peptides in vivo, providing a strategy to prevent accumulation of toxic polyQ peptides. Similarly, the role of alternative degradation pathways, such as autophagy, in clearance of polyQ aggregates can be investigated. Our approach may also be useful in screening for compounds that affect aggregation and decrease toxicity. We expect that the outcome of such studies using this tool, which expresses polyQ peptides in living cells, holds true for all expanded polyQ disorders.

Ub was generated by PCR from GFP-Ub (Dantuma et al., 2006) with forward primer 5′-CCCGAGCTCAGATGCAGATCTTCGTGAAG-3′ and reverse primer 5′-CTCGGGCCCTCACCCACCTCTGAGACGG-3′ and ligated into EGFP-C1 (Clonetech). The resulting construct GFP-Ub was again generated by PCR with forward primer 5′-CGCGGATCCATGGTGAGCAAGGGCGAG-3′ and a reverse primer 5′-CGGGAATTCCTGCAGCCCACCTCTGAGACGGAG-3′, and ligated into Ub-x-GFP-Q16/65/112 (Verhoef et al., 2002) where the Ub-x-GFP insert was replaced by GFP-Ub, resulting in GFP-Ub-Q16/Q65/Q122. This procedure was required to remove the restriction site PstI present between GFP and Ub, since PstI was also required for Ub-polyQ ligation. The use of restriction sites required the presence of some flanking amino acids, resulting in an N-terminal Leu residue and a Glu-Thr-Ser-Pro-Arg sequence at the C-terminus. GFP was exchanged for mRFP to generate the different RFP-Ub-polyQ fusions. The alternative polyQ peptide lengths of Q33 and Q48 were generated by re-transformation of GFP-Ub-Q65, leading to altered polyQ lengths. Q16-GFP was generated by inserting a Q16 repeat (derived from Ub-M-GFP-Q16) in front of GFP. Htt exon-1 was kindly provided by Ron Kopito (Stanford University, Stanford, CA), Atx3 by Henry Paulson (University of Iowa, Iowa, IA), AR by Paul Taylor (St. Jude Children's Research Hospital, Memphis, TN), GFP-Ub, RFP-Ub, Ub-M-GFP-polyQ (used to express GFP-polyQ) and β7-RFP by Nico Dantuma (Karolinska Institutet, Stockholm, Sweden), Hsp70-GFP by Harm Kampinga (UMC Groningen, The Netherlands), TBP1 by Rick Morimoto (Northwestern University, Evanston, IL) and QBP1-CFP by Yoshitaka Nagai (Osaka University Graduate School of Medicine, Osaka, Japan).

Human embryonic kidney cells (HEK293T) and Mel JuSo fibroblast cells where cultured in Iscove's modified Dulbecco's medium (IMDM; Gibco) supplemented with 10% FCS and penicillin-streptomycin-L-glutamine. The cells where transiently transfected with Fugene6 (Roche) and analyzed at the indicated times after transfection. Mouse STHdh^+/+(Q7) cells (kindly provided by Marcy MacDonald) (Trettel et al., 2000) and N2A neuroblastoma cells were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% FCS and penicillin-streptomycin-L-glutamine. Neuronal cells where transiently transfected with Lipofectamine 2000 (Invitrogen). Mouse STHdh^+/+(Q7) cells were incubated at 32°C. For toxicity measurements, N2A cells were analyzed by FACS LSRII for GFP fluorescence 24 or 48 hours after transfection, and the percentage of GFP-positive cells was quantified.

Cytosolic extracts were generated by lysing cells with 0.1% Triton X-100 for 30 minutes on ice, and the supernatant was used after spinning down the lysate. 20 μg of cytosolic protein lysates were separated by 18% SDS-PAGE and transferred to Protan nitrocellulose membranes. Membranes were blocked in 5% dry milk in TBS containing 0.3% Tween, and probed with 1:1000 anti-GFP (Molecular Probes), 1:100 anti-Ub (Sigma) or the anti-polyglutamine 1C2 (MAB1574, Millipore). Polyclonal horseradish peroxidase (HRP)-conjugated secondary antibodies, anti-rabbit (Sigma) or anti-mouse (DAKO) were used at a 1:10,000 dilution to detect the primary antibodies in conjunction with Lumi-lightPLUS western blotting substrate (Roche). Preparation of SDS-soluble and SDS-insoluble protein fractions was described previously (Carra et al., 2008). Briefly, cells were trypsinized, homogenized, and heated for 10 minutes at 99°C in sample buffer (70 mM Tris pH 6.8, 1.5% SDS, 20% glycerol) supplemented with 50 mM dithiothreitol (DTT) 72 hours after transfection. Cell lysates were centrifuged for at least 30 minutes at 20,800 g at room temperature. Bromophenol blue (0.05%) was added to the supernatants, which were the SDS-soluble fraction. The pellets, the SDS-insoluble fractions, were dissolved in 100% formic acid, incubated 30 minutes at 37°C, lyophilized overnight in a speed vac (Eppendorf), and resuspended in a 0.25 volume of sample buffer containing 0.05% Bromophenol Blue. Samples were separated on either 18% SDS-PAGE (anti-polyQ), or 12.5% SDS-PAGE (anti-GFP) and further treated as for western blottings.

HEK293T cells were transfected with the indicated constructs and the percentages of aggregates were scored using an inverted fluorescence microscope (Leica DMR). For imaging, Mel Juso cells were transiently transfected with the indicated constructs and images where obtained using a confocal microscope (Leica SP2) with a ×63 objective. Note that some images show `over-exposed' fluorescent aggregates in order to visualize non-sequestered, cytoplasmic staining. For immunostaining, Mel Juso cells were fixed with 4% paraformaldehyde and permeabilized using 0.1% Triton X-100 in PBS containing 1% FCS, and stained with the primary antibodies 1C2 or MW1 (Ko et al., 2001) (1:1000), followed by goat anti-mouse Cy3 labelling (Jackson ImmunoResearch Laboratories). The MW1 antibody developed by Jan Ko, Susan Ou and Paul Patterson (Ko et al., 2001) was obtained from the Developmental Studies Hybridoma Bank under the auspices of the NICHD and maintained by the University of Iowa. For endogenous Hsp70 labelling, Mel Juso cells were stained for Hsp70/Hsc70 (Calbiochem, 1:200) followed by anti-mouse Alexa Fluor 633 (Invitrogen). For electron microscopy, Mel Juso cells were embedded in situ. Preceding fixation, cells were washed briefly in 20 mM PBS (pH 7.4). Cells were fixed with a mixture of 4% paraformaldehyde, 1% glutaraldehyde in 0.1 M phosphate buffer (pH 7.4) for 60 minutes. They were then washed in distilled water, osmicated for 60 minutes in 1% OsO₄ in water, washed again in distilled water, dehydrated through a series of ethanol baths and embedded in LX-112. After polymerization, the plastic was removed and small parts of the Epon block containing the cells were prepared for ultra-thin sectioning. Ultra-thin sections were cut, collected on formvar-coated grids and stained with uranyl acetate and lead citrate. Sections were examined with a Philips EM-420 electron microscope.

Filter retardation assays were performed as described previously (Wanker et al., 1999). Briefly, 72 hours after transfection, HEK293T cells were lysed for 30 minutes on ice in Nondinet P-40 (NP-40) buffer [100 mM Tris-HCl, pH 7.5, 300 mM NaCl, 2% NP-40, 10 mM EDTA, pH 8.0], supplemented with complete mini protease inhibitor cocktail (Roche) and phosphatase inhibitor cocktail (Sigma). After centrifugation for 15 minutes at 20,800 g at 4°C, cell pellets were resuspended in benzonase buffer (1 mM MgCl₂, 50 mM Tris-HCl; pH 8.0) and incubated for 1 hour at 37°C with 250 U benzonase (Merck). Reactions were stopped by adding 2× termination buffer (40 mM EDTA, 4% SDS, 100 mM DTT). Aliquots of 30 μg protein extract were diluted into 2% SDS buffer (2% SDS, 150 mM NaCl, 10 mM Tris pH 8.0) and filtered through a 0.2 μm cellulose acetate membrane (Schleicher and Schuell) pre-equilibrated in 2% SDS buffer. Filters were washed twice with 0.1% SDS buffer (0.1% SDS, 150 mM NaCl, 10 mM Tris pH 8.0) and subsequently blocked in 5% nonfat milk (Protifar Plus, Nutricia) in TBS. Captured aggregates were detected by incubation with 1C2 antibody and further treated as for western blotting. Alternatively, GFP fluorescence of trapped aggregates was analysed by LAS3000.