Expansion of CAG trinucleotide repeats coding for polyglutamine in unrelated proteins causes at least nine late-onset progressive neurodegenerative disorders, including Huntington's disease and a number of spinocerebellar ataxias. Expanded polyglutamine provokes a dominant gain-of-function neurotoxicity, regardless of the specific protein context within which it resides. Nevertheless, the protein context does modulate polyglutamine toxicity, as evidenced by the distinct clinical and pathological features of the various disorders. Importantly, polyglutamine toxicity might derive from its ability to aggregate. Indeed, aggregation probably underlies some defining attributes of the polyglutamine disorders, such as their late onset, progressive nature, and the dependence of onset age on polyglutamine length. However, the central role of aggregation in polyglutamine pathogenesis has been challenged by several studies, which instead argued that the soluble form of the disease proteins is responsible for neuronal damage. Thus, the question whether polyglutamine aggregates are deleterious, harmless or protective remains the most passionately disputed issue in the study of these diseases. In this review, we attempt to reconcile some of these controversies.
Polyglutamine expansion in unrelated proteins is responsible for at least nine late-onset progressive neurodegenerative disorders, including Huntington's disease (HD; huntingtin), spinal and bulbar muscular atrophy (SBMA; androgen receptor), dentatorubral and pallidoluysian atrophy (DRPLA; atrophin-1), and a number of spinocerebellar ataxias (SCA; ataxins) (1). Numerous observations have established that the polyglutamine repeat by itself plays a central role in the pathogenesis of these diseases, although its effects are strongly modulated by the protein context within which it resides. The discovery in 1997 of insoluble depositions of mutant huntingtin in neurons of a transgenic model of HD suggested that polyglutamine toxicity might derive from its ability to form aggregates (2). This realization sparked an intense inquiry into the mechanisms by which polyglutamine aggregates wreak havoc in cells. Nevertheless, soon afterwards, new findings began to question the supposed causality of aggregation in the pathogenesis of polyglutamine diseases. Among these, the poor overlap between neurons with visible inclusions and neurons actually degenerating in each disorder indicated that aggregates were neither necessary nor sufficient to provoke neuronal damage. In addition, polyglutamine toxicity could be experimentally dissociated from its aggregation in several cellular and animal model systems. Consequently, polyglutamine aggregates are now frequently viewed either as irrelevant by-products, or as structures that protect cells by sequestering the soluble form of polyglutamine proteins, which in turn is considered the principal culprit of toxicity. Nevertheless, the debate concerning the role of aggregation in polyglutamine toxicity remains as fervent as ever, and is regularly invigorated by novel findings. The purpose of this review is to offer a critical summary of the ongoing discussion, and to advocate the importance of aggregation in the pathogenesis of polyglutamine disorders.
A unifying pathological feature of polyglutamine disorders is the presence of microscopically discernible inclusions of the mutant proteins in the nucleus or cytoplasm of affected neurons (Fig. 1) (3–9). Existence of such inclusions indicated that expanded polyglutamine acquired a novel tendency to aggregate, which could be central to pathogenesis of these diseases. Polyglutamine aggregation has been reproduced in vitro, using either a fragment of mutant huntingtin or synthetic polyglutamine peptides (10–12). These studies defined both the aggregation kinetics and the biochemical and structural properties of the resulting aggregates. The aggregation displays kinetics of nucleated-growth polymerization, with a prolonged lag-phase required to form an aggregation nucleus, followed by a fast extension phase during which additional polyglutamine monomers rapidly join the growing aggregate (12,13). The aggregates consist of β-sheet-rich fibrils aligned side-by-side to form ribbon-like structures, and exhibit several defining features of amyloid, such as binding of thioflavine T, congo red birefringence, and reactivity with a generic anti-amyloid antibody (11,14).
Obviously, the propensity of expanded polyglutamine to aggregate must derive from its ability to self-interact. To reveal the molecular details of this interaction, researchers used molecular modelling, which yielded a number of plausible structures (10,15–17). The best-known of these structures—the ‘polar zipper’ put forward by Max Perutz—envisioned polyglutamine tracts forming extended anti-parallel β-strands, held together by an extensive network of hydrogen bonds between both the main-chain and side-chain amides (10). Perutz recognized that the ability of polyglutamine's amide groups to H-bond with solvent water would destabilize the inter-amide H-bonds between the interacting polyglutamine chains. However, he argued that entropic forces would make self-interaction increasingly favourable for longer polyglutamine repeats, thus explaining their tendency to aggregate (18). The ‘polar zipper’ model offered a helpful approximation of the interactions involved in polyglutamine aggregation. Yet, the precise organization of polyglutamine molecules within an aggregate is still largely unknown and remains a subject of intense scrutiny (17).
IS POLYGLUTAMINE AGGREGATION PATHOGENIC?
In light of the ongoing effort to uncover the molecular details of polyglutamine aggregation, one must inquire whether aggregation constitutes the initial step in the cascade of events leading to neuronal dysfunction and demise in polyglutamine disorders. To evaluate the importance of different mechanisms in initiating the polyglutamine pathogenesis, several authors formulated criteria for a relevant pathogenic ‘trigger’: this must explain: (i) why symptoms only occur above a threshold of approximately 36–40 glutamines; (ii) why above this threshold increasing polyglutamine size leads to earlier appearance and augmented severity of symptoms; (iii) how expanded polyglutamine causes a dominant gain-of-function; (iv) why only neurons are affected despite the mutant proteins being expressed ubiquitously; and (v) why distinct neuronal populations are lost in each of these diseases (19,20). In the following, we examine whether and how polyglutamine aggregation could underlie these observations.
Analysis of patient's material established that polyglutamine expansion beyond the pathological threshold of 36–40 glutamines leads to concurrent appearance of both disease phenotype and inclusions of the mutant proteins. Similarly, early in vitro analyses suggested that polyglutamine aggregation only occurred above the pathological threshold (11). This indicated that repeats longer than 36–40 glutamines acquired a qualitatively novel—and probably pathogenic—propensity to aggregate.
To offer a structural explanation for this propensity, Perutz speculated that above the pathological threshold polyglutamine monomers in solution would favour a conformational change from a statistical coil to a stable intramolecular β-structure, perhaps a β-hairpin or a β-helix (18,21). In turn, cohesion of an adequate number of such ‘misfolded’ polyglutamine monomers would generate a nucleus and set off aggregation (22). However, this hypothesis was recently challenged by structural studies, which did not detect any novel β-conformation for soluble expanded polyglutamine (12,14,23,24). Rather, soluble polyglutamine existed as a random coil and only converted into β-sheet in a template-aided process coincident with aggregation (12,14). Intriguingly, the nucleus that sparked this aggregation consisted of a single ‘misfolded’ polyglutamine monomer (25). Hence, any assembly of two or more polyglutamine monomers, in which they assume a distinctive (though as yet unknown) structural organization typical of fibrillar aggregates, should be considered an aggregate.
Moreover, Chen and colleagues (12,14) demonstrated that peptides much shorter than 36 glutamines were also able to aggregate spontaneously, although their aggregation kinetics showed considerably delayed nucleation and less rapid extension compared to longer peptides. Therefore, the pathological threshold should not be viewed as a qualitative divide between ‘aggregating’ and ‘non-aggregating’ polyglutamine lengths. Rather, only polyglutamine repeats above this threshold will eventually form aggregates in human tissues, because the time required for nucleation and growth of aggregates of shorter repeats would probably extend beyond the average human lifespan, perhaps up to hundreds or thousands of years (25).
Genotype–phenotype correlations in the various polyglutamine disorders revealed that longer polyglutamine repeats provoke earlier appearance of the disease symptoms, more generalized pathology and faster progression. Importantly, several attributes of polyglutamine aggregation could account for these observations. First, in vitro analysis of aggregation kinetics demonstrated that above the pathological threshold longer polyglutamine tracts display more rapid nucleation (12–14). Hence, aggregation will take place sooner in patients with longer repeats, perhaps explaining the earlier onset of symptoms in these patients. Second, longer polyglutamine also exhibits lower ‘critical concentration’, that is the minimal concentration of monomer required for aggregation to happen (12–14). Therefore, more expanded polyglutamine will form aggregates in a broader range of neurons—both those with high expression of the mutant protein, as well as the ones with lower expression—well in agreement with the more widespread clinical and pathological alterations seen in patients with larger expansions. Finally, the lower critical concentration might also explain the accelerated progression and increased phenotypic severity in patients with longer repeats, since in these patients a greater proportion of neurons in vulnerable brain regions would develop aggregates within a given time, leading to more pronounced functional decline.
‘Multiple nucleation’ hypothesis.
When discussing the relation between aggregation and the above disease parameters, i.e. onset age, rate of progression and phenotypic severity, we must ask whether aggregation results from a single nucleation event, or whether multiple nuclei (and aggregates) arise within a cell. In vitro, nucleation is the rate-limiting event, whereas elongation of the emerging aggregate advances rapidly until monomers are largely depleted from the solution, thereby precluding formation of additional nuclei. Likewise, in cellular systems massively overexpressing polyglutamine-expanded proteins, extensive aggregation ensues within minutes after a nucleus has formed (26). In contrast, the rate of aggregate elongation in a patient's neurons might be markedly reduced due to: (a) low cellular concentration of the polyglutamine disease proteins; (b) decreased mobility of the polyglutamine-containing proteins due to their interactions with other proteins or due to molecular crowding within the cell; (c) poor accessibility of the polyglutamine tract within the context of the proteins; (d) compartmentalization of emerging aggregates into cellular structures specialized to sequester misfolded and aggregated proteins, such as the perinuclear aggresome (27–29) and the nuclear promyelocytic leukemia domains (30–34); and (e) because of continuous refolding and degradation of the aggregated proteins by molecular chaperones and the proteasome (35). Therefore, in neurons a single nucleation event might not set off large-scale aggregation, but instead generate a small, slowly growing aggregate, leaving the bulk of the polyglutamine-expanded protein soluble. This would increase the likelihood of additional nuclei subsequently arising within the same neuron, resulting in a single cell harbouring plentiful minute aggregates—or microaggregates. Ultimately, such microaggregates would coalesce together, perhaps with active assistance from the cell (29,36–38), to produce larger structures discernible by light microscopy or macroaggregates.
Two important predictions follow from this line of reasoning. First, cells would contain abundant polyglutamine aggregates long before inclusions become visible by light microscopy. In fact, formation of such large inclusions would not increase the total amount of aggregated polyglutamine in the cells. Second, because of its faster nucleation, longer polyglutamine would yield more numerous aggregates within a single neuron, thereby increasing the total cellular load of aggregated polyglutamine. This might in part underlie the more rapid progression and augmented phenotypic severity in patients with more expanded repeats.
Unfortunately, experimental data on the process of aggregation in vivo is sparse. Yet some observations in transgenic models of polyglutamine diseases do favour the ‘multiple nucleation’ scenario. In these models, the polyglutamine proteins gradually accumulate inside neuronal nuclei, producing a ‘diffuse’ signal that intensifies with advancing age of the animal, later evolves to include a number of small punctuate structures, and finally gives rise to a single inclusion with no residual diffuse staining (Fig. 2) (39–44). Obviously, the ‘diffuse’ nuclear staining in these mice might in fact result from the presence of numerous minute aggregates. Indeed, such microaggregates were detected using immuno-electron microscopy in a transgenic model of Huntington's disease (39). Moreover, in SCA7 knock-in mice, insoluble mutant ataxin-7 could be detected long before inclusions became evident on light microscopy (45). In contrast, another study asserted that the ‘diffuse’ signal in an HD knock-in model originated from soluble polyglutamine, because aggregates—as assayed by the filter-retardation technique—only appeared after inclusions became visible (40). However, it is debatable whether the filter-retardation assay—which distinguishes polyglutamine aggregates based on their resistance to solubilization by strong detergents and on their size exceeding 0.2 µm (11)—would also identify microaggregates, since these might perhaps disintegrate in detergents or remain below the 0.2 µm cut-off. Clearly, a systematic temporal analysis of the intermediate forms of polyglutamine aggregates in vivo is needed to resolve these controversies.
In conclusion, polyglutamine aggregation can explain the existence of pathological threshold, as well as the progressivity of the disease phenotype with increasing repeat lengths above this threshold. However, how do polyglutamine aggregates induce neuronal dysfunction and death?
Mechanisms of aggregate toxicity
Loss of function.
The simplest manner in which aggregation could initiate cellular dysfunction would be by sequestering the respective mutant proteins away from their usual localization, thereby compromising their function and leading to haplo-insufficiency. Moreover, if a certain disease protein was able to self-interact—as is the case for ataxin-1 and ataxin-3—then aggregates of the mutant protein might also sequester the wild-type counterpart, causing a dominant negative loss-of-function. However, genetic evidence compellingly rejects a central role for haplo-insufficiency or dominant negative loss-of-function in pathogenesis of polyglutamine disorders. For instance, individuals hemizygous for the SCA1 or HD genes do not present with any of the neurological characteristics of spinocerebellar ataxia or Huntington's disease, respectively. Likewise, males lacking the X-linked androgen receptor gene do not develop SBMA. Finally, SCA1 nullizygous mice show altered hippocampal function, but no signs of cerebellar ataxia (46).
Nevertheless, loss-of-function might partly contribute to neuronal demise in some polyglutamine diseases. Most notably, wild-type huntingtin encourages neuronal survival by blocking the activation of apoptotic cell-death cascades (47–50), and by inducing the production of brain derived neurotrophic factor (BDNF) (51). Intriguingly, apoptotic neurons are noticed both in human HD tissue (52–54) and in transgenic mice expressing mutant huntingtin (55). Moreover, BDNF levels are markedly reduced in the striatum and cortex of another HD transgenic model (51). Together this suggests that loss of huntingtin's normal function might aggravate neurodegeneration in HD. Indeed, when huntingtin expression is abolished in brains of adult mice, the animals manifest with a progressive neurological phenotype and neuropathological alterations reminiscent of mice expressing polyglutamine-expanded huntingtin (56).
Numerous observations argue convincingly that expanded polyglutamine by itself elicits dominant gain-of-function neurotoxicity, which is common to all polyglutamine disorders. First, with the exception of SBMA, all polyglutamine diseases are autosomal dominant. SBMA itself is X-linked recessive, because heterozygous females are protected against the neurotoxic effects of polyglutamine-expanded androgen receptor by their low levels of circulating androgens (57,58). Second, while all polyglutamine disorders present as late-onset progressive neurodegenerations, the various disease proteins are not homologous outside of their polyglutamine tracts, and probably do not participate in related neuronal processes. Third, expression of the mutant proteins in transgenic mice harbouring two intact copies of their respective wild-type homologues still provokes neuronal dysfunction and degeneration (39,42–44,55,57,59–62). Moreover, several disease proteins also induce neurodegeneration in Drosophila and C. elegans, which lack functional homologues of these proteins (58,63–66). Finally, expanded polyglutamine induces neuronal toxicity in transgenic models even if carried by proteins entirely unrelated to the known disease proteins (67), or when expressed without any adjoining protein sequences (68–71).
How can aggregation trigger such dominant toxicity of expanded polyglutamine? In answering this dilemma, one discovery was particularly revealing, namely, that aggregates in patients' brains are not composed solely of the respective disease proteins, but rather attract a plethora of additional cellular proteins. This suggested that sequestration into aggregates might deplete select cellular factors away from their usual localization, thereby compromising their function and leading to toxicity; this proposal became widely known as the ‘sequestration’ hypothesis. Hence, what kinds of proteins are sequestered into polyglutamine aggregates, and which molecular forces mediate their interaction with the aggregated disease proteins?
Proteins harbouring short, non-pathogenic glutamine repeats might associate with polyglutamine aggregates via homotypic glutamine–glutamine interactions. Indeed, glutamine peptides as short as 15 residues are readily incorporated into pre-formed aggregates of longer polyglutamine in vitro (12), and proteins with short glutamine tracts co-localize into polyglutamine aggregates in both transfected cells and animal models (72–74). Intriguingly, high incidence of such short glutamine repeats in transcriptional regulators, together with the preferential localization of this class of proteins inside the nucleus—the principal site of aggregation in polyglutamine diseases—would strongly encourage the sequestration of transcriptional regulators into polyglutamine aggregates. In fact, a variety of glutamine-rich transcriptional regulators, including the cAMP response element binding protein binding protein (CBP) (73,75–77), TATA-binding protein (TBP) (72), TBP-associated factor (TAFII130) (78), and specificity protein 1 (Sp1) (78), do localize into inclusions of polyglutamine disease proteins both in cellular and animal models, and in patients' tissue. Sequestration of these factors into polyglutamine aggregates might underlie the characteristic alterations in gene expression patterns observed in transgenic models of several polyglutamine disorders (79–81).
In particular, recruitment into aggregates of the transcriptional co-activator CBP, which carries 18 consecutive glutamines, might greatly exacerbate neuronal demise in polyglutamine disorders. CBP cooperates with the transcription factor cAMP response element binding protein (CREB) in activating the expression of cAMP-responsive genes. Importantly, CREB-dependent transcription protects neurons in response to stressful stimuli such as hypoxia, and mediates the survival-promoting effect of neurotrophins, including BDNF (82). In agreement with a vital role for CREB-dependent transcription in neuronal survival, mice with postnatal disruption of both Creb and its homolog Crem display progressive neurodegeneration in the hippocampus and the striatum (83).
Numerous observations indicate that CREB-dependent transcription is compromised in polyglutamine disorders: (a) expression levels of cAMP-responsive genes are diminished in early Huntington's disease brains, as well as in cellular and animal models of HD (79,84–87); (b) hippocampal long-term potentiation (LTP), which too requires functional CREB, is impaired in HD and SCA1 mice (60,88,89); (c) cultured neuronal cells overexpressing truncated mutant huntingtin respond poorly to neurotrophins (90,91); and (d) cAMP and forskolin, both of which enhance CREB transactivation, rescue polyglutamine toxicity in transfected cells (91,92).
Likewise, a causal role for CBP sequestration in impairing CREB-dependent transcription in polyglutamine diseases is supported by ample evidence: (a) CBP is found in aggregates of several distinct polyglutamine proteins in cellular and animal models and in patients' brains (73,75–77); (b) levels of soluble CBP are diminished in HD brains (77); and (c) CBP overexpression alleviates polyglutamine toxicity in neuronal cells (76).
Nevertheless, it should be kept in mind that depletion of CBP would probably compromise the function of other transcription factors besides CREB (93). Conversely, CREB-dependent transcription might be thwarted by sequestration of other transcriptional co-activators, namely TAFII130, into polyglutamine aggregates (78). Finally, not all data substantiate the role of CBP sequestration in polyglutamine pathogenesis: for instance, one study failed to confirm the co-localization of CBP into nuclear inclusions in three different transgenic models of HD (94).
Molecular chaperones and the proteasome.
Molecular chaperones and the proteasome are attracted into polyglutamine aggregates by recognizing the misfolded nature of the aggregated proteins (95,96). While molecular chaperones associate with the aggregated proteins only transiently and move freely into and out of the aggregates (74), the proteasome becomes engaged in recurring attempts to unfold and degrade the ubiquitinated aggregated proteins (97). Redistribution of the proteasome into aggregates might limit its availability in the cell, leading to diminished clearance and harmful accumulation of other misfolded or damaged cellular proteins (97,98). This would eventually activate cellular stress response and induce apoptosis.
However, the proteasome is in fact able to entirely degrade existing polyglutamine aggregates in neurons after expression of the soluble mutant protein is discontinued (35). This observation rejects the notion that the proteasome is irreversibly trapped and inactivated inside the aggregates. Therefore, it is unlikely that cells would suffer a significant decrease of proteasome activity, unless they accumulate exceptionally high quantities of aggregated polyglutamine. Yet, if a cell amasses enough aggregated polyglutamine to reduce proteasome's availability by just a little, this would initiate a self-reinforcing cascade: less free proteasome means slower degradation of the soluble polyglutamine protein, leading to enhanced aggregation and further decline in the availability of free proteasome (97).
Interaction partners of the disease proteins.
Additional proteins might be recruited into aggregates through interactions with the non-polyglutamine segments of the individual disease proteins. Consistent with this prediction, composition of aggregates of mutant ataxin-3 closely parallels the recognized interactions of the wild-type protein (34). Importantly, because the various disease proteins differ markedly outside of their polyglutamine tracts, their aggregates will probably recruit distinct sets of interacting proteins. This will result in different cellular functions being affected in the various disorders, thereby adding to the phenotypic differences. Nevertheless, while a substantial number of interacting partners have been identified for most of the disease proteins, it is often unknown whether their sequestration into aggregates exacerbates cellular dysfunction.
Substrates of transglutaminase.
Finally, a range of cellular proteins might become covalently linked to polyglutamine aggregates by the action of tissue transglutaminase, which catalyses formation of cross-links between glutamine and lysine residues. This binding would immobilize and perhaps inactivate the proteins in question. Yet, the importance of tissue transglutaminase-mediated cross-linking in polyglutamine pathogenesis remains controversial (99).
In summary, sequestration probably constitutes a key mechanism of polyglutamine aggregate toxicity. Intriguingly, sequestration can explain both the dominant gain-of-function toxicity of expanded polyglutamine per se, as well as the modifying effects of the protein context. Importantly, while the common view is that recruitment of certain proteins into aggregates would result in a loss of-function of these proteins, there might exist other mechanisms by which sequestration affects cellular functions. Most notably, recruitment of caspase-8 and -10 into polyglutamine aggregates leads to anomalous auto-activation of these proteases and initiation of cell death cascades (100).
Toxicity of polyglutamine aggregates in the cytoplasm.
An alternative mechanism how polyglutamine aggregates, when localized into neuronal processes, might disrupt neuronal function is by physically obstructing axonal trafficking, leading to axonal degeneration (101–103). However, this mechanism would only apply to those disorders in which the mutant proteins aggregate both in the nucleus and in the cytoplasm, namely HD and SCA2 (4,8,104).
Aggregation can also explain why neurons constitute the primary target of polyglutamine toxicity, despite the different mutant proteins being expressed at similar levels in both neural and peripheral tissues. For example, unlike in neurons, which are post-mitotic, polyglutamine (micro)aggregates in cycling cells would be distributed between daughter cells upon each cell division, thereby reducing the actual load of aggregated polyglutamine. Moreover, when the nuclear membrane disintegrates in dividing cells, polyglutamine (micro)aggregates relocate from the nucleus into the cytoplasm, where they might elicit less toxicity (91,105). Consistent with these notions, polyglutamine toxicity in transfected cells is greatly enhanced when their cycling is arrested by mitotic blockers or differentiation-promoting agents (91,106).
Perhaps the most perplexing aspect of polyglutamine disorders is that only certain neuronal populations degenerate in each condition, despite most disease proteins being expressed widely across the brain. Therefore, we must inquire whether aggregation could explain the restricted pattern of neuronal loss in these diseases.
In every disorder, aggregates might preferentially arise in select neuronal groups due to variation in quantity, subcellular localization, turnover, proteolytic cleavage, or protein interactions of the respective mutant proteins. While it might appear that only selective presence of aggregates within susceptible neuronal types would support a causal role for aggregation in pathogenesis, such expectation is an immense oversimplification. First, without detailed understanding of the mechanisms leading to aggregate toxicity, one must not assume that distinct neurons are equally vulnerable to polyglutamine aggregates. For instance, although aggregates of all polyglutamine disease proteins will sequester additional cellular proteins, both the identity of such proteins and the consequences of their recruitment into aggregates might differ between distinct neuronal types, as well as between the various disease proteins. Second, post-mortem neuropathological examination only describes the final outcome of a disease process that usually takes many years. Therefore, it is hard to establish a clear temporary relationship between the appearance of aggregates and development of disease symptoms. Moreover, while aggregates are found in both vulnerable and spared neurons in post-mortem material, they might have formed considerably sooner, and thus exert their toxicity for longer, in affected neurons. Third, neuronal populations not harbouring any aggregates might still degenerate due to non-cell autonomous, trans-neuronal mechanism. For example, degeneration of striatal neurons in Huntington's disease results in part from diminished import of BDNF from cortical neurons (51). Likewise, in a transgenic model of SCA7, cerebellar Purkinje cells display marked functional and morphological deterioration, despite the absence of mutant ataxin-7 expression in these neurons (107). Finally, methodological constraints also confound the relation between aggregation and toxicity. For instance, distribution of aggregates across the CNS is typically assessed using light microscopy, which reliably detects large inclusions, but overlooks the smaller microaggregates. Similarly, existing biochemical assays of aggregation, such as the filter-retardation assay, might underestimate the incidence of small aggregates. Yet, such minute aggregates are probably as harmful, if not more, as the large inclusions, because their greater surface-to-volume ratio makes them particularly adept at sequestering other proteins (12). In fact, formation of large inclusions might even represent a protective effort by the cell to minimize the surface of the aggregates in contact with cellular environment (29,36–38).
In light of the above arguments, it is little surprising that the observed distribution of aggregates does not entirely overlap with the recognized patterns of neuronal loss in the various disorders. For example, while in SBMA (7), SCA1 (5) and SCA3/MJD (3) nuclear inclusions (NI) are found predominantly in affected brain regions, in SCA7 they form in both affected and spared areas (9,108). Moreover, in Huntington's disease, NI are rare in the striatum, the region most vulnerable to degeneration, and are in fact confined to neuronal types that are spared in HD (104,109). However, aggregates of mutant huntingtin are abundant in cortical neurons projecting to the striatum, suggesting that striatal degeneration might result in part from dysfunction of these neurons (104). Finally, recent morphometric analysis of SCA1 and SCA3/MJD brains demonstrated that within one affected brain region (pons), the size and shape of the nuclei was more preserved in neurons harbouring visible NI than in neurons without NI (110,111). This suggests that formation of such large inclusions might indeed be beneficial.
In conclusion, post-mortem evaluation of patient's material can neither decisively confirm nor reject a causal role for polyglutamine aggregation in pathogenesis. Rather, this conundrum must be addressed in transgenic and knock-in models of polyglutamine diseases, which offer a unique chance to dissect the temporal relationship between aggregation and pathogenesis (see below).
EXPERIMENTAL MODELS OF AGGREGATE TOXICITY
As detailed above, the recognized features of polyglutamine aggregation conform well to many of the criteria for a relevant trigger of polyglutamine pathogenesis. Nevertheless, the supposed causality of aggregation in pathogenesis had to be thoroughly tested in experimental model systems of polyglutamine diseases. A large number of cellular and animal models of polyglutamine toxicity have been examined in trying to substantiate the role of aggregation in toxicity. In essence, this amounted to inspecting (a) whether aggregation presaged the onset of cellular dysfunction and death; (b) whether cells bearing aggregates were more severely affected than cells with soluble polyglutamine; and (c) whether treatments that modulate polyglutamine aggregation could alter its toxicity in predictable ways.
Studies in transfected cells generated a flood of often-contradictory data, both in favour of and in conflict with aggregation constituting the toxic insult. This inconsistency reflects the enormous variability of experimental designs employed in the cell culture studies. Indeed, a huge collection of different mutant proteins or their fragments, carrying anywhere between 40 and 300 glutamines has been expressed in a wide variety of cell types. In addition, most studies massively overexpress the mutant proteins, resulting in aggregation and cell death within days rather than years as seen in the diseases. Therefore, the molecular pathways triggering death of polyglutamine-transfected cells, such as activation of stress response and induction of apoptosis, might not correctly approximate the processes leading to the slowly advancing dysfunction and degeneration of neurons in the human diseases.
An additional drawback of cellular models is that in order to obtain aggregates one must first overexpress the soluble mutant proteins. However, because overexpression, aggregation, and cell death follow each other in rapid succession in transfected cells, the relative contributions of the soluble versus aggregated polyglutamine to cellular malfunction are hard to tell apart. To overcome this limitation, Yang and colleagues introduced pre-aggregated polyglutamine peptides into nuclei or cytoplasm of cultured cells (112). Intriguingly, the aggregates caused cell death when targeted into the nuclei, implicating a nucleus-specific mechanism of polyglutamine aggregate toxicity in this system. Moreover, aggregates of a non-polyglutamine protein were harmless when introduced into the nuclei, arguing that the toxicity of polyglutamine aggregates did not result from mechanical disruption of nuclear processes by the aggregates, but rather derived from polyglutamine's ability to interact with select nuclear proteins.
Fortunately, animal models—particularly those expressing truncated fragments of disease proteins with highly elongated polyglutamine repeats—offered more solid evidence in favour of aggregation triggering neuronal dysfunction. Representing this collection of mouse models, the well-known R6/2 mice express an N-terminal portion of huntingtin with over 140 glutamines (113). These mice develop abundant nuclear aggregates long before the onset of motor dysfunction and the appearance of neurological abnormalities (2,114). Moreover, despite marked reduction in brain size, the animals do not show extensive neuronal loss until after a prolonged interval of neuronal dysfunction (115). Hence, aggregated polyglutamine provokes progressive neuronal dysfunction, reflected as phenotypic deterioration, which after a delay leads to demise of the affected neurons. While the R6/2 phenotype might be somewhat unique, since the mice develop aggregates both in nuclei and in neuropil (116), the above scenario also applies to transgenic models harbouring exclusively nuclear aggregates (42,44,59,69,117,118). A further argument supporting the role of aggregation in polyglutamine toxicity came from a transgenic model with regulated expression of mutant huntingtin fragment (35). In these mice, down-regulation of transgene expression leads to disappearance of nuclear inclusions, which is then followed by marked improvements of motor performance.
Suppression of aggregation
Perhaps the most compelling evidence in favour of aggregation being harmful is that reducing aggregation alleviates toxicity. In particular, increased cellular levels of certain molecular chaperones, especially Hsp70 and Hsp40, concomitantly suppress both aggregation and toxicity of expanded polyglutamine in numerous cell culture systems (95,96,119–121), as well as in Drosophila (63,70,122,123) and mouse models (124,125). Intriguingly, in several of these models, overexpression of molecular chaperones does not in fact reduce nor delay the formation of visible inclusions (122,124), but instead markedly increases the solubility of expanded polyglutamine within these inclusions (123). Similarly, addition of molecular chaperones to polyglutamine in vitro prevents appearance of detergent-resistant fibrillar aggregates, and in its place favours formation of detergent-soluble amorphous aggregates (126). Polyglutamine molecules in such amorphous aggregates will probably not generate the highly organized, repetitive network of intermolecular interactions typical of fibrils. Rather, they might be held together by irregular alignment of only restricted portions of the polyglutamine tracts. Nevertheless, it is unclear why the latter molecular organization of polyglutamine aggregates would be less harmful to cells. Perhaps, the less compact arrangement of amorphous aggregates would make them more amenable to proteasome-mediated degradation (127). Alternatively, such aggregates might be less adept at irreversibly sequestering other cellular proteins.
Besides molecular chaperones, other aggregation-opposing agents—such as chemical chaperones (128), synthetic or recombinant peptides (26,129,130), intracellular antibodies (131,132), and certain chemicals (133–135)—also attenuate polyglutamine toxicity, further substantiating the central role of aggregation in pathogenesis. However, many of these agents, including molecular chaperones themselves (98), prevent aggregation by (transiently) binding to the soluble form of polyglutamine-expanded proteins. Therefore, it is not unequivocally ruled out that these agents might in fact reduce toxicity by neutralizing the potential harmful effects of soluble polyglutamine.
Perhaps the most tantalizing results were obtained using the azo-dye congo red. This chemical not only blocks oligomerization and aggregation of expanded polyglutamine, but can also disintegrate existing aggregates. Intriguingly, congo red prevented much of the toxicity of expanded polyglutamine both in transfected cells and in the R6/2 transgenic model of Huntington's disease (135). In the R6/2 mice administration of congo red well after the formation of nuclear inclusions and appearance of neurological symptoms resulted in significant improvement of motor performance and extended lifespan. Importantly, these phenotypic improvements coincided with extensive clearance of existing polyglutamine aggregates in the basal ganglia of these mice. Thus, the use of congo red (or its analogues) represents a promising avenue in treatment of polyglutamine diseases.
IS SOLUBLE POLYGLUTAMINE PATHOGENIC?
Importantly, with the advent of transgenic mouse models of polyglutamine disorders, it became increasingly apparent that aggregation might not constitute the sole mechanism of polyglutamine neurotoxicity. For example, mice expressing full-length mutant huntingtin exhibit neurological phenotype and marked neurodegeneration, despite showing little aggregation (39,55). Likewise, SCA1 knock-in mice display degeneration of cerebellar Purkinje cells, although these neurons do not develop prominent aggregates (60). Extending these findings, several influential studies argued that aggregation might not be the chief mechanism of polyglutamine toxicity.
First, Klement and colleagues demonstrated that two distinct transgenic models of SCA1—the first one expressing full-length mutant ataxin-1, while the other one carrying an internal deletion within mutant ataxin-1—showed a comparable ataxic phenotype, although only the mice expressing the full-length protein developed nuclear aggregates in cerebellar Purkinje neurons (136). However, subsequent analysis revealed that ataxia in the transgenic mice lacking Purkinje cell inclusions was in fact non-progressive (137). Hence, while aggregation of ataxin-1 might not be necessary to elicit functional alterations in Purkinje cells, it seems to underlie the progressive worsening of the phenotype. Alternatively, Perutz (138) suggested that the internal deletion within ataxin-1 might have denatured the protein, leading to toxicity entirely unrelated to the actual mechanisms of SCA1 pathogenesis. Unfortunately, the study could not refute this possibility, since control mice expressing internally deleted, wild-type ataxin-1 were not generated.
Second, Cummings and co-workers (37) examined the effects of mutant ataxin-1 expression in Purkinje neurons of transgenic mice deficient in the ubiquitin ligase E6-AP. Surprisingly, absence of E6-AP delayed the appearance of nuclear inclusions, while at the same time accelerating both the onset of ataxia and the morphological deterioration of Purkinje cells. This indicates that the large ataxin-1 inclusions might not represent the most harmful form of the mutant protein. However, the study did not reject a causal role for ataxin-1 aggregation in SCA1 pathogenesis, since the effect of E6-AP deficiency on solubility of mutant ataxin-1 was not assessed. Indeed, it remains possible that ubiquitination of ataxin-1 microaggregates would promote their active sequestration into a larger structure, a nuclear inclusion, thereby thwarting their toxicity. Consistent with this idea, inclusions of most polyglutamine proteins arise within distinct subnuclear compartments, the promyelocytic leukaemia domains, which function as centres of sequestration and proteasome-mediated degradation of misfolded and aggregated nuclear proteins (30).
Third, Saudou and others (36) observed that interference with ubiquitin conjugation suppressed aggregation, but enhanced the toxicity of truncated mutant huntingtin in transfected primary striatal neurons. However, noting that disruption of ubiquitin conjugation is in itself harmful to these neurons, Rubinsztein (139) argued that the combined toxicities of polyglutamine-expanded huntingtin and of ubiquitin conjugation deficiency might have resulted in accelerated death and detachment of cells containing aggregates, thus causing an underestimation of aggregation frequency.
In conclusion, the above studies did not compellingly reject the involvement of aggregation in polyglutamine toxicity. Rather, they suggested that the large, microscopically discernible inclusions do not constitute the only toxic species of expanded polyglutamine. Quite the opposite, formation of these inclusions might in part represent an active effort by the cell to sequester smaller, intermediate forms of aggregated polyglutamine, thereby reducing their contact with cellular environment and thwarting their toxicity. Yet, the outcomes of these studies have too often been regarded as evidence against the role of aggregation in polyglutamine pathogenesis. We like to emphasize that both soluble and aggregated polyglutamine might harm neurons via non-exclusive mechanisms. Therefore, any evidence supporting the toxicity of one form of polyglutamine should not be automatically regarded as evidence against toxicity of the other form.
Mechanisms of toxicity of soluble expanded polyglutamine
When contemplating how the soluble form of the various disease proteins could elicit neuronal toxicity, it is crucial to distinguish the toxic effects of soluble expanded polyglutamine per se from the impact of the polyglutamine expansion on the metabolism, interactions and functions of the respective disease proteins. Here we only address how soluble expanded polyglutamine by itself might be toxic to neurons.
Three novel attributes of monomeric expanded polyglutamine might trigger its toxicity: (i) increased strength of its interactions with certain cellular components; (ii) ability to form ion channels across cellular membranes; and (iii) inhibitory effect on the proteasome.
Structural studies have established that polyglutamine in solution exists as a random coil, and that elongation beyond the pathological threshold does not coincide with it assuming any novel, potentially pathogenic conformation (12,14,23,24). Yet, despite not having altered conformation, soluble expanded polyglutamine does acquire a new property—a dramatically improved binding to proteins (or protein complexes) containing two or more regions with affinity for glutamine tracts (23). If such binding disrupted the function of critical cellular components, this would clarify why expanded polyglutamine becomes cytotoxic.
Which cellular processes might be compromised by binding of soluble expanded polyglutamine to their molecular components? Intriguingly, the transient association of transcriptional regulators into transcriptional complexes is commonly facilitated by interactions between glutamine-rich regions within these proteins. Hence, it is plausible that soluble expanded polyglutamine would bind to transcriptional mediators harboring glutamine tracts, disrupt the organization of transcriptional complexes, and lead to transcriptional dysregulation. Indeed, soluble polyglutamine has been shown to interact directly with several glutamine-rich transcriptional regulators—including CBP (77), TAFII130 (78,140), and Sp1 (140,141)—and this interaction is enhanced by expansion of the polyglutamine tract into the pathological range. Reflecting this interaction, expanded polyglutamine impairs transcriptional processes mediated by some of these proteins (77,78,140,141). Importantly, many of the transcriptional regulators that interact with soluble polyglutamine also bind to polyglutamine aggregates. Thus, expanded polyglutamine might be a double-edged sword, disturbing transcription whether soluble or aggregated.
Another potentially harmful interaction takes place between expanded polyglutamine tracts and the acetyltransferase domains of several histone acetylases, including CBP, p300 and P/CAF (142). The function of these enzymes is to maintain chromatin in a trancriptionally active state through acetylation of histones. Importantly, expanded polyglutamine inhibits histone acetylases and impairs histone acetylation, thereby probably compromising transcription. Indeed, the role of defective histone acetylation in polyglutamine pathogenesis has been substantiated by observation that histone deacetylase inhibitors, which preserve the acetylation of histones by blocking the action of histone deacetylases, aleviate neurodegeneration in Drosophila and cellular models of polyglutamine toxicity (142,143).
Toxic channel hypothesis.
Monoi and colleagues (144) reported that expanded polyglutamine in a helical conformation can form H+- and K+-permeable membrane channels. Importantly, such channels would readily dissipate ionic concentration gradients existing across the various cellular membranes. In particular, diminution of the mitochondrial H+ gradient would have detrimental effects on cellular energy metabolism and calcium homeostasis. In agreement with this proposal, mitochondria isolated from lymphoblasts of Huntington's disease patients, and from the brain of an HD transgenic model display slightly reduced resting potential and faster depolarization in response to added Ca2+ (145,146). This Ca2+-dependent depolarization occurs via opening of the mitochondrial permeability transition pore, which might lead to activation of programmed cell death. Intriguingly, similar alterations are also observed using wild-type mitochondria, to which purified expanded polyglutamine is added, arguing that they result from a direct impact of the polyglutamine tracts on mitochondrial function (146).
Importantly, while expanded polyglutamine would provoke only minimal alterations in the baseline performance of mitochondria, it would severely frustrate their ability to buffer out excessive cytoplasmic Ca2+. Because excitatory aminoacid such as glutamate increase the influx of Ca2+ into the cytoplasm, this predicts that neurons harboring expanded polyglutamine would be exceptionally vulnerable to excitotoxicity, i.e. neuronal death resulting from chronic exposure to excitatory aminoacid (147). In agreement with this expectation, excitotoxicity plays a major role in degeneration of striatal medium-spiny neurons in Huntington's disease (147).
As already mentioned, soluble expanded polyglutamine does not acquire any novel, compactly folded conformation, suggesting that it should be accessible to unfolding and degradation by the proteasome. Indeed, when soluble expanded polyglutamine is targeted for proteasome-mediated degradation in transfected cells, it is cleared as efficiently as normal-size polyglutamine (148) (A. Michalik and C. Van Broeckhoven, manuscript in preparation). Moreover, several agents that prevent polyglutamine aggregation, such as molecular chaperones, polyglutamine-binding peptides and congo red, also stimulate the clearance of the mutant proteins, arguing that it is the oligomers and microaggregates, rather than the soluble expanded polyglutamine, that resist degradation (26,127,135). In contrast, polyglutamine-expanded ataxin-1 and ataxin-7 do exhibit reduced turnover, although the exact mechanism underlying these observation is unknown (37,149).
Intriguingly, Goellner and Rechsteiner (150) argued that proteasome-mediated degradation of expanded polyglutamine might be compromised specifically in neurons. These authors observed that the proteasome complex capped with the REGγ regulatory particle is virtually unable to cleave after glutamine residues. Because the REGγ particle is markedly enriched in neuronal nuclei, and because peptides larger than 35 amino acids cannot diffuse out of the proteasome, they speculated that the proteasome in neurons would gradually be ‘choked up’ by undigested glutamine peptides remaining within its catalytic cavity. In line with this hypothesis, Ding et al. (151) reported that expression of expanded polyglutamine in stably transfected neuroblastoma cells impaired the ability of the proteasome to respond properly to additional cellular stress. Since very few cells contained aggregates in this experiment, it is likely that soluble expanded polyglutamine exerted a chronic inhibitory effect on the proteasome.
One intriguing question related to the toxicity of soluble expanded polyglutamine is how the constant presence of mutant polyglutamine induces the late-onset, progressive phenotype characteristic of the polyglutamine diseases. Some effects of soluble polyglutamine, such as the impairment of mitochondrial Ca2+ metabolism and inhibition of the proteasome, will probably be cumulative. Hence, neurons will have to accumulate a critical amount of damage to these systems before this will result in functional deterioration or cell death. Thus, clinical phenotype will only ensue after a delay, and will be progressive. Moreover, longer polyglutamine repeats would speed up the accumulation of such cellular injury, thereby causing earlier onset and faster progression.
Alternatively, other effects of soluble expanded polyglutamine, for example its aberrant binding to certain cellular components, might alter neuronal homeostasis in a steady manner, thereby increasing the chance of spontaneous activation of cell death pathways (152). In this so-called ‘one-hit’ scenario, neurons exhibit a constant probability of death throughout the lifetime, while the phenotype appears only after a given proportion of vulnerable neurons have been lost. Thus, a persistent effect of expanded polyglutamine on neuronal survival translates into a late-onset, progressive phenotype. Importantly, this scenario might apply to Huntington's chorea, in which the striatum already suffers a considerable loss of medium-spiny neurons before the onset of symptoms. In this case, the persistent effect of mutant huntingtin that renders medium-spiny neurons vulnerable to (excitotoxic) cell death might be the chronic overactivation of a specific subtype of NMDA glutamate receptors (153).
In summary, we outlined some of the key aspects of polyglutamine pathogenesis (Fig. 3). We greatly favour a role for aggregation in polyglutamine toxicity, not least because it explains many facets of the polyglutamine diseases as seen in patients. Indeed, in our opinion there exists no data conclusively refuting the involvement of aggregation in polyglutamine toxicity. At most, the available data implies that increased levels of soluble expanded polyglutamine might be cytotoxic as well, perhaps impinging on the very same cellular processes as the aggregated form.
In our review, we did not address the mechanisms by which polyglutamine expansion within the disease proteins might compromise the cellular functions in which they normally participate. For instance, reduced turnover and ensuing higher levels of polyglutamine-expanded ataxin-1 and ataxin-7 will probably influence their functioning in neurons (37,149). Likewise, nuclear build-up of mutant huntingtin fragments—which directly inhibit histone acetylases—will probably impair trancription (142). In addition, polyglutamine expansion might alter the strength of interactions between the various disease proteins and their normal binding partners. Reduced binding would mimic a partial loss-of-function effect: for instance, polyglutamine expansion in huntingtin weakens its interaction with Hip-1, resulting in increased levels of free Hip-1, which then activates caspase-8 dependent cell-death cascade (49). On the contrary, the consequences of increased binding might be harder to predict. For example, mutant huntingtin provokes overactivation of a specific subclass of NMDA glutamate receptors, leading to excitotoxicity in the striatal spiny projection neurons (153). Moreover, increased interaction of mutant ataxin-1 with the polyglutamine-binding protein 1 (PQBP-1) results in inactivation of RNA polymerase II (154). However, it is unclear to what degree these mechanisms of pathogenesis are shared among the diverse polyglutamine disease. In fact, by being specific to distinct disease proteins, such mechanisms might constitute the main determinants of the phenotypic differences between the disorders.
The authors acknowledge Jean-Jacques Martin and Jean-Louis Mandel for providing Figures 1 and 2, respectively. The research on polyglutamine diseases in the author's laboratory of Molecular Genetics is funded by the Fund for Scientific Research Flanders, Belgium.
To whom correspondence should be addressed at: Department of Molecular Genetics VIB8, University of Antwerp, Universiteitsplein 1, B-2610 Antwerpen, Belgium. Tel: +32 38202601; Fax: +32 38202541; Email: email@example.com