Physical chemistry of polyglutamine: Intriguing tales of a monotonous sequence

Abstract

Polyglutamine sequences of unknown normal function are present in a significant number of proteins, and their repeat expansion is associated with a number of genetic neurodegenerative diseases. Polyglutamine solution structure and properties are not only important because of the normal and abnormal biology associated with these sequences, but also because they represent an interesting case of a biologically relevant homopolymer. As the common thread in the expanded polyglutamine repeat diseases, it is important to understand the structure and properties of simple polyglutamine sequences. At the same time, experience has shown that sequences attached to polyglutamine, whether in artificial constructs or in disease proteins, can influence structure and properties. The two major contenders for the molecular source of the neurotoxicity implicit in polyglutamine expansion within disease proteins are a populated toxic conformation in the monomer ensemble and a toxic aggregated species. This review summarizes experimental and computational studies on the solution structure and aggregation properties of both simple and complex polyglutamine sequences, and their repeat-length dependence. As a representative of complex polyglutamine proteins, the behavior of huntingtin N-terminal fragments, such as exon-1, receives special attention.

Although functionally obscure polyglutamine (polyQ) sequences are found in a variety of proteins ^1–3, many of which are transcription factors ^1,4, and although aggregation-prone plant storage proteins rich in Gln are also well-known ^{5, 6}, it is unlikely that polyQ sequences would ever have become a major focus of research were it not for their pathogenic role in proteins associated with a series of inherited, devastating neurodegenerative diseases ^{7, 8}. At least 10 ^{8, 9} of these expanded CAG repeat diseases have been described, including the most prevalent and most well-known condition, Huntington’s disease (HD) ¹⁰. All of them are triggered by inheritance of a polyQ sequence with a repeat length longer than normal, where normal is in the 20–35 range for all but one of these diseases ⁸. All of them feature the accumulation in particular brain regions of polyQ-containing neuronal aggregates, and all of them appear to be primarily gain-of-function diseases ⁸. These and other data have suggested that expanded polyQ repeat proteins are defective in some aspect of protein folding, such as a lowered barrier to misfolding within the monomer and/or to aggregate formation.

Because of the great interest in understanding the molecular mechanisms of these diseases and in developing therapies, the biophysical properties of polyQ sequences have received considerable scrutiny in the nearly 20 years since the pathogenic role of polyQ expansion was first realized ¹¹. While the exact folded and aggregated states of the toxic species in these diseases are yet to be worked out, a wealth of important biophysical data has been accumulated. In addition to any implications for mechanisms and therapeutic strategies of neurodegenerative diseases, some of these data have far-reaching implications for general protein solution behavior and aggregation. Experimental biophysical studies on polyQ solution structure and aggregation can also provide data to support computational analyses that have potential relevance to polyQ pathology as well as to fundamental polymer physics. PolyQ studies have provided important information on the solution structure of intrinsically disordered proteins, and on the all-important mechanisms by which amyloid growth is initiated.

This review focuses on in vitro experimental studies of polyQ solution and aggregate structure and on the mechanisms of spontaneous aggregation and its inhibition. Selected results from the wealth of in silico and in vivo studies are also discussed. The reader is referred to a recent review for more thorough coverage of biological aspects ¹⁰.

Polyglutamine Solution Structure

The polyQ monomer in water

The observation of ubiquitinylated, polyQ containing aggregates in neuronal nuclei of HD patients ¹² and animal models ¹³, in addition to other factors, led directly to the hypothesis that expanded CAG repeat diseases should be included in the growing family of human pathologies involving protein misfolding and aggregation ¹⁴. Since globular protein aggregation ¹⁵, including amyloid formation ^16–18, is understood to be associated with breakdowns in protein conformational integrity, speculation on the nature and source of expanded polyQ gain-of-function has included not only a proposed leading role for protein aggregation but also the related, but distinct, hypothesis of an important, toxic misfolded state within the monomer ensemble ⁷ that might also, coincidentally, mediate aggregation. But what is the evidence for the existence of a distinct, highly populated folded state in either benign or pathogenic polyQ sequences? In this section I review what is known about populated folding states of polyQ monomers.

Initial evidence for a predominantly disordered structure for polyQ in solution came from CD analysis of solutions of short polyQ monomers ¹⁹. Subsequent CD studies on pathological length polyQ sequences, either chemically synthesized ²⁰ or biosynthetic ²¹, showed that even long polyQ sequences are largely disordered, and that there is no obvious difference in secondary structure between long and short polyQ. NMR studies ²² subsequently confirmed the disordered nature of polyQ monomers of all repeat lengths, and supported earlier CD studies ²³ that showed the absence of detectible, conformationally distinct intermediates in the coil to sheet transition that occurs when polyQs grow into amyloid. X-ray diffraction structures of crystals of a sequence-modified huntingtin N-terminal fragment containing a short polyQ insert showed the intervening polyQ sequence to be in a variety of conformations, including α-helix ²⁴. CD analysis shows that a simple K₂Q₄₀K₂sequence has about 10% α-helix configurations at 35 °C and takes on an additional 10% on cooling to 5 °C ²⁵. The overall picture from solution phase experiments is that polyQ sequences, regardless of repeat length, are largely disordered chain with transient elements of regular secondary structure. This impression is supported by a number of computational analyses starting from unstructured chains that, while not agreeing completely with each other, suggest that both short or long polyQ sequences generally exhibit mixtures in various proportions of disordered structure, α-helix, β-sheet, PP_IIhelix, and β-turns ^26–32.

There is an additional aspect to the solution structure of the polyQ sequence that likely plays an important role in its behavior – the tendency of the sequence to collapse, in spite of its apparent hydrophilic nature, in aqueous solution. This behavior has been demonstrated in several biophysical studies. A fluorescence correlation spectroscopy (FCS) analysis of a collection of labeled polyQ molecules of differing repeat lengths ³³ demonstrated two important points. First, that polyQ peptides behave in a simple aqueous buffer as a polymer in a poor solvent. That is, the polyQ chain in water is neither fully extended in a statistical coil conformation, as it would be in a good solvent, nor is it somewhat less extended, as occurs under “theta” conditions just short of the point where long-range interactions begin to contribute to polymer conformations. Rather, polyQ in water behaves as a polymer in a bad solvent, in which solvent water is excluded from the collapsed polymer chain. Second, that the compactness of the molecule does not change appreciably as polyQ repeat length increases from Q₁₅ to Q₅₄, that is, through the typical disease threshold of ~35. The relatively compact structure of polyQ in solution was confirmed in two FRET studies ^{32, 34}. Atomic force stretching experiments on polyQ sequences embedded within artificial flanking sequences are also consistent with a collapsed structure in contact with water, suggesting a very high barrier to the unfolding or extension of hydrated polyQ sequences ³⁵. These results are also consistent with computer simulations that predict a collapsed coil structure for monomeric polyQ, likely held together by H-bonding between side chain and main chain amide groups, in preference to H-bonding to water ^{27, 29, 36, 37}. This internal H-bonded structure could well play a role in the expected highly inefficient formation of β-hairpin ³⁰ or other regular structures ³⁷ that might be implicated in aggregation nucleation (see below).

Is there a “misfolded“ state of polyQ?

Inconsistent with the above data on a polyQ repeat-length independent disordered state are antibody binding data that have been interpreted to suggest the existence of populated alternative conformations of monomeric polyQ that become more highly populated at repeat lengths above the typical disease risk threshold of ~35–45. Trottier and colleagues showed that the anti-polyQ monoclonal antibody 1C2 binds considerably better, in Western blots, to longer polyQ sequences compared with short versions ³⁸. This led to the idea that 1C2 recognizes a specific polyQ conformation that is much more highly populated in long polyQ sequences than in short, and therefore to the idea that 1C2 recognizes a toxic conformation populated only in long polyQs. An alternative explanation emerged, however, when the Bjorkman group showed that a similar MAb, MW1, owes its preferential binding to long polyQ peptides to a “linear lattice” effect of the polyQ sequence ³⁹. According to their analysis, the affinity of MW1 for a polyQ sequence is enhanced according to the number of individual epitopes in the polyQ chain, even though each individual epitope might be short (as was, in fact, subsequently shown by an X-ray crystal structure of the complex ⁴⁰).

More recently, another antibody with a preference for long polyQ sequences, 3B5H10, has been described and its selective binding also interpreted as evidence for a toxic folded conformation in long polyQ sequences ⁴¹. While this remains formally possible, antibody binding does not prove that a particular conformation is populated in the absence of the antibody. The thermodynamics of complex formation allows antibodies to recruit and enrich kinetically accessible conformations even if they are only minimally populated in the absence of the antibody ⁴². Conformational recruitment is an especially attractive possibility when the target molecule is an intrinsically disordered protein like polyQ that is presumably capable of undergoing coupled folding and binding ^{43, 44} to more structurally organized binding partners.

Although there is no convincing evidence for a populated, pre-existing conformation in expanded polyQ sequences that might be considered a “toxic conformation”, it remains possible, nonetheless, that monomers may yet prove to be the toxic species. This could happen if the toxic event (a binding interaction, for example, to some cellular target) is only favorable with a polyQ sequence of a particular length. Models for such binding phenomena might be (a) the linear lattice ³⁹ type of polydentate interaction, (b) a binding cavity (such as found in some molecular chaperones or proteasomes) satisfied by the compact coil state of a sufficiently large polyQ molecule, or (c) a templated, coupled folding and binding event ⁴³ (such as aggregate elongation, for example) with a thermodynamic requirement for a certain amount of buried surface area that can only be supplied by expanded polyQs. Thus there are a number of mechanisms by which one would expect to observe a correlation between cell toxicity and cellular levels of monomeric expanded polyQs, as has been recently observed ⁴¹. Such correlative studies do not prove the existence of a populated, toxic, misfolded conformation in the monomer ensemble, however, and in fact merely formally indicate a pool of molecules that are capable of undergoing a toxic binding event.

Solution structure of complex polyQ sequences

In all expanded CAG repeat proteins, the polyQ sequence is embedded within the protein and therefore has both N- and C-terminal flanking sequences. Sequence algorithms make somewhat contrasting predictions of secondary structure of both the polyQ insert and surrounding sequence, with one program predicting disorder ⁴⁵ and another predicting the ability to assemble into coiled coils ⁴⁶. Besides the structural details of a particular peptide in solution, one would also like to know how the flanking sequences influence polyQ structure and properties, and how the polyQ sequence (and its expansion) influences surrounding structure. There are significant technical limitations on addressing such questions experimentally with intrinsically unstructured proteins, requiring specialized spectroscopic approaches ^{44, 47}. Perhaps because of this, a number of groups have looked at polyQ disease protein structure issues through computational approaches.

There has been a special emphasis on sequences representing the N-terminal 3% of the huntingtin sequence (Fig. 1), which includes the polyQ segment as well as an N-terminal sequence (htt^NT) of 17 amino acids and a C-terminal segment rich in proline residues. A hint that flanking sequence can influence polyQ conformation in the monomer ensemble comes from the ability a C-terminal P₁₀ sequence to slow the nucleated growth of polyQ amyloid and to abrogate the ability of a temperature decrease to enhance α-helix formation in a polyQ sequence ²⁵. A series of crystal structures of a significantly altered htt N-terminal fragment containing a short polyQ repeat show the htt^NTto fold into an α-helix, the C-terminal proline rich sequence to explore PP_IIstructures, and the intervening polyQ to occupy a variety of structures, including α-helix ²⁴. The extent to which crystal packing forces dictate any of these results is not clear but certainly must be considered possible, especially for a peptide that is disordered in solution.

Figure 1.

Structure of huntingtin. The 3144 amino acid long human huntingtin protein (numbering based on a Q₂₃ repeat length) showing the N-terminal exon-1 segment and other regions of interest. Boxed in red are three domains rich in ~40 amino acid long HEAT repeats implicated in protein-protein interactions ¹⁰, as identified by a neural network analysis ¹⁹¹. Dark green, htt^NT segment; orange, polyQ; purple, proline-rich; cyan, phosphorylation and acetylation sites; magenta, in vivo caspase and calpain cleavage sites suggesting a disordered region; blue, nuclear export sequence; light green, nuclear import sequence. Based on information reviewed in reference ¹⁰.

The solution structure of a protein includes its native oligomerization state, and oligomerization can influence secondary structure in profound ways. Recent evidence suggests that htt N-terminal fragments are capable of reversible tetramer formation, almost certainly mediated by the formation of an α-helical bundle centered in the htt^NT segment. The htt^NT sequence in isolation as a monomer is disordered ⁴⁸, exhibiting a slight tendency toward α-helix ^{48, 49}. Recent work shows that the α-helix content of the htt^NT segment depends on concentration, and that at concentrations of 0.5 mM or higher, the peptide is essentially fully helical ⁵⁰ (Fig. 2a). These results, coupled with an analytical ultracentrifugation (AUC) analysis suggesting that htt^NT and htt^NTQ₁₀K₂ molecules exist at modest concentrations as an equilibrium of monomer, tetramer, octomer, and perhaps higher oligomers ⁵⁰ (Fig. 2b) suggests that htt N-terminal fragments are capable of facile oligomerization through α-helix bundle formation via their htt^NT segments. Consistent with the AUC data, various fluorescence techniques previously indicated an early assembly in cells of GFP-tagged htt exon-1 proteins into small multimers ^{51, 52}. Small multimers were not observed in striatal neurons expressing htt exon-1, however ⁴¹. The htt^NT sequence has been proposed to be an important regulatory appendage of htt, mediating – with or without post-translational modifications – a variety of targeting, trafficking, and clearance operations in the cell ^{49, 53–58}. In this context, it seems possible that reversible tetramer formation may be a means by which solution exposure of the promiscuous htt^NT sequence can be itself regulated. Through this or other means, it is possible that tetramer formation by expanded polyQ versions of htt N-terminal fragments may somehow play a direct role in toxicity. The implications of htt^NT-mediated oligomer formation for the aggregation nucleation mechanism of htt N-terminal fragments is discussed later.

Figure 2.

Htt^NT-mediated oligomerization. (a) Concentration dependence of the CD spectra of htt^NTQ: green, 1.3 mM; blue, 0.67 mM; orange, 0.33 mM; black, 0.15 mM; red, 0.10 mM obtained by dilution from a 0.76 mM solution. (b) Sedimentation velocity analysis of 50 μM htt^NTQ₁₀K₂ at 20 °C. Top panel, sedimentation data; middle panel, residuals from the global fit; bottom panel, c(s) profile with peaks labeled (in kDa). Data from reference ⁵⁰.

Two groups have computationally studied the htt^NT segment in isolation. Kelley et al. found a high tendency for all or part of the htt^NTsegment to form α-helix structure ⁵⁹. Rossetti et al. found a number of conformational families which, except for a small fraction of condensed coil, are dominated by extended, solvated structures with helical segments ⁶⁰. In general, these findings are in agreement with experimental data that show monomeric htt^NT to be disordered at low concentrations ⁴⁸, and to form helical tetramers at higher concentrations ⁵⁰.

Other groups have conducted simulations on various htt N-terminal fragments. Consistent with the polyproline effect on aggregation mentioned above, Lakhani et al. found that the polyproline region of htt exon-1 decreases the ability of polyQ to engage β-structure elements within the monomer ³¹. This group also found that polyQ expansion leads to increased β-structure in the htt^NT segment ³¹. In contrast, Williamson et al. found some α-helix tendency within the htt^NT segment of htt^NT-Q_N sequences (i.e., without the proline domain), and a tendency to increased disorder in htt^NT as polyQ repeat length increases ⁶¹. Dlugosz and Trylska found an intriguing repeat length dependent conformational change within an htt N-terminal fragment ⁶². For both long and short polyQ sequence versions, both the htt^NTand the polyQ were largely α-helical. In the short polyQ molecule, the PP_II helix of the proline-rich tail tends to dock to the htt^NT helix, while in the long polyQ version these elements do not interact and are therefore more solvated ⁶². The variety of simulation results reflected by the above discussion presumably results from the use of different simulation methods ⁶².

A number of model studies have been conducted on polyQ fusion proteins, and attempts made to interpret CD or FTIR data to deduce structural consequences. Placing polyQ inserts between secondary structural elements within a folded protein to various extents can destabilize and/or alter the folding or local structure of the host protein ^63–67. In a roughly repeat-dependent fashion, polyQ inserts tend to explore disordered and/or β-structure ^{65, 67}. Interestingly, polyQ sequences fused to some intact folded domains tend to acquire additional α-helical character ^67,68, although other protein hosts don’t seem to detectibly affect polyQ structural preferences ²¹. In general, polyQ adjacent to a folded domain does not appreciably affect native structure and/or folding stability of the host protein ^{66, 67}.

What is the normal role of polyQ in protein structure and function?

In some settings, polyQ repeat length correlates inversely with protein activity ^{69, 70}, suggesting the possibility that repeat expansion diseases, while predominantly gain-of-function ⁸, might also possess some loss-of-function character in some cases ⁷¹. As a general rule, however, it is not clear that polyQ segments in proteins actually have important, specific functions. Rather, they might simply have developed over the course of evolution at protein sites where polyQ tracts are tolerated, undergoing relatively cost-free expansion at these sites due to the replicational instability ⁷² of the CAG repeat sequence. An example is shown in Table 1, a line-up of the polyQ-proximal sequences of various TATA box binding proteins (TBPs) over a wide range of evolution. While the sequences surrounding the polyQ are well-conserved, the polyQ repeat length itself ranges from 4 to 38 (Table 1). Such sequence expansion in an otherwise highly conserved protein suggests a linker between, or adjacent to, structural and/or functional units. At such sites normal protein function might be largely retained even when the polyQ sequence expands to a pathological repeat length. Such a scenario is roughly compatible with the results from model system studies discussed in the preceding paragraph.

Table 1.

PolyQ Repeat Lengths in TATA Box Binding Proteins (TBPs) of Various Species^a

Source - GenBank	N-terminal to polyQ	QN	C-terminal to polyQ
Xenopus NP_001089038.1	-TYGTGLTPQPVQTTNSLSILEEQQR	4	--------TQQSTLQQGNQG-SGQTPQ
Zebrafish AAH55549.1	-PYGTGLTPQPVQNSNSLSLLEEQQR	6	---------AASQQQGGMVGGSGQTPQ-
Viper Q92146.1	-PYGTGLTPQPAQSTNSLSILEEQQR	6	-------AAAQQSTSQPTQAPSGQTPQ-
Chicken NP_990434.1	-PYGTGLTPQPVQSTNSLSILEEQQR	6	-------AAQSSTSQQATQGTSGQTPQ-
Rat AAH81939.1	-PYGTGLTPQPVQNTNSLSILEEQQR	15	AVATAAASVQQSTSQQPTQGASGQTPQ-
Cow AAI13309.1	-PYGTGLTPQPIQNTNSLSILEEQQR	19	---AAVAAVQQSTSQQATQGPSGQTPQ-
Rabbit XP_002723543.1	-PYGTGLTPQPIQNTNSLSILEEQQR	20	AVAAAAAAVQQSTSQQAAQGVSGQTPQ-
Chimpanzee NP_001098077.1	-PYGTGLTPQPIQNTNSLSILEEQQR	35	--AVAAAAVQQSTSQQATQGTSGQAPQ-
Human AAI10342.1	-PYGTGLTPQPIQNTNSLSILEEQQR	38	--AVAAAAVQQSTSQQATQGTSGQAPQ-

^aAlignments fron Kalign (EMBL).

Aggregation of Simple Polyglutamine Sequences

Because the only apparent common element of the proteins associated with the 10 expanded CAG repeat diseases is the polyQ sequence, the solution and aggregation behavior of simple polyQ sequences has been seen by several groups as a reasonable starting point in studying this disease family ^{20, 23, 25, 33, 34, 45, 73–79}. One common non-Gln element of these sequences has been in inclusion of flanking charged residues ^{19, 73, 80, 81} which bestow on these peptides a degree of kinetic (i.e., short term) solubility that allows their solution properties, including nucleation of aggregation, to be studied. The presence of flanking charged residues appear to be quite important; a polyQ sequence lacking flanking charged residues is reasonably well-behaved during disaggregation, but precipitates immediately upon adjustment to PBS conditions ⁸¹. While it might be argued that a polyQ peptide lacking flanking charged residues is a more appropriate setting in which to study the properties of polyQ sequences in a larger protein context, polyQ fused to a number of disease protein flanking sequences retains the good kinetic solubility of K₂Q_NK₂ molecules, in contrast to the poor solution properties of polyQ lacking flanking residues, ^{25, 45}. While charged flanking residues probably introduce some qualitative rate differences, most of the experiments described in this section employ flanking charged residues to make physical experiments and comparisons feasible.

Soon after the observation that expanded polyQ aggregates accumulate in neurons ^{12, 13}, in vitro studies on recombinantly produced exon-1 fragments demonstrated robust aggregation into amyloid-like fragments that is both repeat-length dependent and concentration-dependent ⁸². Subsequently, in vitro studies on chemically synthesized polyQ peptides in the K₂Q_NK₂ background showed similar repeat length dependent aggregation ²⁰ into ordered, amyloid-like ⁸¹ aggregates. The repeat length dependence was observed in both a spontaneous aggregation reaction dependent on both nucleation and elongation properties and in a fibril growth assay requiring only elongation of existing seeds ²⁰. A nearly identical repeat length dependence was subsequently demonstrated in a C. elegans model ⁸³. The microtiter-based elongation assay also showed that amyloid-like aggregates of one polyQ repeat length are able to seed elongation of other polyQ repeat lengths ²⁰, supporting recruitment-based mechanisms for aggregate toxicity ⁸⁴.

Nucleation of simple polyQ aggregation

The concentration dependence of an aggregating system is often dictated by the nucleation mechanism ⁸⁵. Studying the concentration dependence not only reveals clues to the nucleation mechanism, it also provides insight into how the rates observed for relatively high, laboratory concentrations might translate into the lower, steady state concentrations in the cell. An initial investigation of the concentration dependence of spontaneous aggregation for relatively long polyQ peptides in the K₂Q_NK₂ series revealed the surprising result of a rather shallow concentration dependence consistent with a critical nucleus (n*) of one ²³ (Fig. 3a). Subsequent studies on shorter polyQ sequences in the same series revealed that n* itself is repeat length dependent ⁴⁵. PolyQ peptides K₂Q₁₈K₂ and K₂Q₂₃K₂ follow concentration dependence consistent with n* ≈ 4, while polyQ peptides of K₂Q₂₄K₂ and K₂Q₂₅K₂exhibit n* ≈ 2. PolyQ peptides of K ₂Q₂₆K₂or longer have n* ≈ 1 ⁴⁵. This discovery put the n* ≈ 1 value from earlier work ²³ into context, suggesting that all simple polyQ peptides aggregate according to a classical nucleated growth polymerization mechanism.

Figure 3.

Nucleation mechanisms. (a) Mechanism for simple polyQ aggregation nucleation by an energetically unfavorable conformation of the condensed coil polyQ monomer, where k₁ and k₋₁ are the forward and reverse rates of nucleus formation, K_n^* is the equilibrium constant for nucleus formation, and k₊ is the second order rate constant for elongation of the nucleus ^{23, 45, 75, 76}; (b) classical thermodynamic model for nucleated growth polymerization (from reference ⁴⁵).

The nucleus for simple polyQ aggregation, regardless of the n* value, can be modeled as a small β-sheet assembly, as expected (although technically not required) for an amyloid growth nucleus. As repeat length increases, folded versions of monomers, themselves containing β-sheet elements, take on enhanced (but still very low) stabilities that allow them to simplify nucleus assembly by requiring the involvement of fewer molecules ⁴⁵. One attractive model for these transiently populated conformations is the β-turn ^{45, 74, 86}. Indeed, one study showed that polyQ peptides containing L-Pro-Gly pairs compatible with β-hairpins aggregate as rapidly as simple polyQ of the same repeat length, and a similar peptide containing D-pro-Gly pairs that encourage β-hairpin formation aggregate faster than simple polyQ ⁷⁴. Both peptides exhibit the same n* ≈ 1 as do unbroken polyQs in this repeat length range, consistent with models for nucleus structure invoking folding into a β-sheet assembly ⁴⁵. A recent study showed a similar rate enhancement for D-Pro-Gly in a short polyQ background ⁷⁹. The conformational details underlying nucleation and aggregation of polyQ sequences containing Pro-Gly interruptions have been examined by computation analyses ^{87, 88}.

The model for nucleation of aggregation by a long polyQ peptide (Fig. 3a) depicts the monomer ensemble as a condensed coil structure, as suggested by FCS measurements ³³, and the nucleus as a specific folded form possessing β-structure. While nucleus formation is therefore depicted as a folding reaction, it is a highly unfavorable folding reaction. This unfavorable folding reaction is a pre-equilibrium, and the nucleus, once it forms, can either collapse back to the condensed coil monomer ensemble or can undergo a productive elongation step by interacting with a condensed coil monomer, forming an elongation intermediate by (presumably) elaborating a β-stand already present in the nucleus. Subsequent folding steps consolidate β-sheet rich, amyloid-like structure in a now-stabilized dimer that is more likely to engage in a new elongation step, and less likely to disintegrate ³⁰. This multi-step elongation process can be thought of as a “dock-and-lock” type of elongation mechanism ⁸⁹ in which newly added monomers organize at the growth point of the template fibril, rather than encounter the growth point in a pre-organized state.

The ideal behavior of the polyQ nucleation mechanism has allowed a number of studies to elucidate details of the mechanism. The mathematical treatment of the mechanism-based kinetic model yields not only n* but also a complex parameter containing K_n^*, the equilibrium constant for nucleus formation, and k₊, the second order rate constant for nucleus elongation ^{23, 85} (Fig. 3a). A value for k₊ for a K₂Q₄₇K₂ peptide of 1.14 × 10⁴ M⁻¹sec⁻¹ was obtained by determining the pseudo-first order kinetics rate constant for a seeded elongation of this peptide, then converting that value into the second order rate constant using experimental estimates of the concentration of fibril growth points ⁷⁵. Using this second order rate constant, a value for K_n^* of 2.6 × 10⁻⁹ was obtained ⁷⁵. ΔG_folding of + 12.2 kcal/mol (compared to the typical negative ΔG_folding for a small protein of −5 to −15 kcal/mol ⁹⁰), illustrates the rarity of the critical nucleus and the low favorability of nucleus formation.

In the classical nucleated growth mechanism, the efficiency with which the ephemeral critical nucleus elongates is a critical component of the overall efficiency of nucleation. Along with K_n^*, the relative rates of nucleus disintegration and nucleus elongation dictate nucleation efficiency (Fig. 3a). Since the rate of nucleus elongation = k₊[nucleus][monomer], this rate can be stimulated, without affecting nucleus formation, by adding relatively high concentrations of short polyQ peptides ⁷⁶. The end result is an increase in nucleation efficiency by short polyQ sequences. This consequence of the promiscuity of polyQ amyloid elongation may have in vivo significance ⁷⁶.

The ability to fit aggregation kinetics to a mechanism-based mathematical model also allows generation of simulated aggregation kinetics, using the parameters from the fit, under conditions where measurement is not possible. Such simulations predict that even very small differences in n* have huge consequences for nucleation efficiency at the low polyQ concentrations of the cell. Thus, projection of aggregation curves for various repeat lengths of K₂Q_NK₂ to 1 nM concentration shows onset of aggregation in the 10–100 year span for K₂Q₃₇K₂ and 10–100 billion years for K₂Q₂₃K₂ ⁴⁵ (Fig. 4).

Figure 4.

Calculated aggregation kinetics curves for 1 nM of various polyQ peptides using parameters derived from nucleation kinetics analyses. From reference ⁴⁵.

Deviations from the classical nucleated growth model

In the classical nucleated growth model, the overall aggregation reaction is depicted as a series of homologous equilibria, each of which involves the binding of a ground state monomer to the species A_n to generate the species A_n+1 ⁸⁵ (Fig. 3b). Initial binding equilibria (starting with the interaction of two monomers to make A₂) are thermodynamically unfavorable, and later equilibria (approaching a limiting, low monomer concentration) are favorable, and the dividing line between these two classes is the critical nucleus A_n^*, the least stable species on the aggregation pathway ⁸⁵. In the context of this formulation of the classical mechanism, it might be argued that a result of n* = 1 is impossible, since, according to that definition, all species on the assembly pathway are oligomeric. However, what if we slightly alter the definitation of the critical nucleus as a consolidation of structure (i.e., regardless of molecularity) with sufficient stability (and hence lifetime) to sometimes undergo a productive elongation encounter? This does not change the mathematical formalism, while easily accommodating the observation of n* = 1. In fact, it is worth noting that the surprising result of n* = 1 for long polyQ peptides is generated by a data treatment based on a classical model which never anticipated this most simple of solutions. In the next section, the classical nucleated growth model is also found to be too narrow to accommodate an aggregation mechanism for complex polyQ sequences, even though it too features the stochastic formation of a critical species required for amyloid growth.

Other mechanistic models

Based on computational modeling that suggests an ability of polyQ peptides to form dimers ³⁷ and higher oligomers ⁹¹ through non-specific interactions, and the observation that such oligomers can form ⁹², including during fibril formation ^{77, 93}, an alternative nucleation model has been proposed in which amyloid nuclei are generated from within these oligomeric intermediates ^{77, 94}. As further support for this model, an analysis showed that it is compatible with previously reported kinetics data, and could account for the non-integer values that sometimes are reported for n* ⁹⁴. Models involving non-amyloid intermediates do have a certain attraction, given the ubiquitous appearance of on- or off-pathway oligomeric intermediates in almost all other amyloid systems ⁹⁵. There are several issues with this alternative model, however, as it is applied to simple polyQ peptides. First, observations of non-amyloid oligomers from simple K₂Q_NK₂ peptides are quite possibly due to artifacts in sample preparation (see discussion in aggregate morphology section). Indeed, careful adherence to disaggregation protocols ^{45, 96} and a variety of tests focused on detecting low levels of oligomers failed to show any evidence for non-amyloid aggregates during amyloid formation by K₂Q_NK₂ peptides ⁴⁵. Furthermore, although fractional n* values are often reported ^{23, 25, 45, 74–76, 78}, there are other theoretical rationalizations for non-integer n* values, and it has also proved technically difficult to determine n* values with accuracies sufficient to show that non-integer values are significantly different than the nearest integer value ⁴⁵. Finally, direct conversion of amorphous aggregates into amyloid-like aggregates by computer simulations has not been reported, nor have observed non-amyloid oligomeric intermediates been successfully incorporated into kinetic models of aggregation. While it remains possible that very low levels of very small oligomers might be missed in experiments, it has not been necessary to invoke such species in order to account for observed nucleation kinetics ⁴⁵.

Interestingly, it is possible that an oligomer-mediated mechanism applies to the aggregation of histidine-interruped polyQ peptides in the acid pH-range (see below). And there is a very clear case of oligomer-mediated amyloid nucleation in the case of htt N-terminal fragment aggregation, as discussed later.

PolyQ sequence interruptions

Between the world of simple, unbroken polyQ sequences, as discussed above, and complex polyQ proteins with extensive flanking sequences, as discussed below, are peptides that consist of mostly polyQ but containing sequence interruptions. The limited analysis of these sequences suggests that their aggregation is qualitatively similar to unbroken polyQ but with altered kinetics. Sequences containing Pro or Pro-Gly interruptions ^{74, 79, 97–99} are discussed elsewhere in this review.

Sequences containing His interruptions are of special interest, since in Spinocerebellar ataxia 1 (SCA1), one of the expanded CAG repeat diseases, repeat expansion is linked to loss of internal His residues. At the protein level, some individuals have expanded polyQ tracts but retain the His residues, and, perhaps because of that, are symptom-free ¹⁰⁰. Analysis showed that such His insertions do not appear to alter the disordered state of monomeric peptides ⁹⁷, and in the neutral pH range do not alter the nucleation mechanism ⁷⁸. However, His interruptions do substantially reduce aggregation kinetics ^{78, 97}. A proposed model for aggregate structure ⁹⁷ in which the His residues are placed in reverse turns was supported by a structural analysis of the aggregates ⁷⁸. Interestingly, His-containing polyQ peptides incubated at pH 6 experience a significantly reduced aggregation rate and exhibit an alternate aggregation reaction profile with clear evidence for a transient non-amyloid intermediate ⁷⁸. Whether this is on- or off-pathway, however, was not determined. These changes are presumably brought about by protonation of the His residues at the lower pH.

More extreme examples of mixed polyQ sequences should also be mentioned. In a dramatic example, the mixed Ala/Gln sequence K₂-(QA)₂₀-K₂, derived from the htt interacting protein CA150, forms amyloid fibrils with approximately the same kinetics as a K₂Q₄₀K₂ sequence ¹⁰¹. Furthermore, amyloid-like aggregates of these two sequences can “cross-seed” the elongation of the other ¹⁰¹, which is typically not a very efficient process for amyloidogenic peptides of widely different sequences ¹⁰². These results are especially interesting given the possible involvement of amyloid formation in some polyAla repeat diseases ¹⁰³. The other mixed sequences discussed in this section, containing His insertions or Pro-Gly insertions, are also very effective in cross-seeding reactions with simple polyQ sequences, suggesting a high degree of structural similarity ^{74, 78}.

Complex Aggregation Mechanisms for Complex PolyQ Sequences

All known expanded CAG repeat disease proteins contain a polyQ sequence surrounded by mixed amino acid sequence peptides. Some of these flanking sequences do not alter the fundamental aggregation nucleation mechanism, although they may affect aggregation rates ^{25, 45}. In other cases, however, flanking sequences can play an enormous role, greatly affecting rates, sometimes by fundamentally changing the aggregation mechanism. This multiplicity of effect is mirrored in studies on artificial polyQ fusion protein model systems. In some cases, producing the polyQ sequence as a fusion protein greatly suppresses amyloid formation, facilitating purification of the protein and allowing control over the start of aggregation via specific protease cleavage sites between fusion partner and polyQ domain ¹⁰⁴. Studies on intact fusion proteins give a range of results. Some fusion proteins produce only non-amyloid aggregates ⁶⁷, some make amyloid via non-amyloid, oligomeric intermediates ^{105, 106} and some produce amyloid fibrils with little or no evidence for non-amyloid intermediates ⁶⁶. For those proteins that feature formation of non-amyloid intermediates, an important question is the extent to which polyQ expansion influences the first step, versus the second step, of the two-step mechanism of amyloid formation.

Ataxin-3

An excellent example of the ability of flanking sequence to dramatically alter the aggregation pathway is the role of the “Josephin” domain of the disease protein ataxin-3 (AT3). The Josephin domain by itself or with a short polyQ extension undergoes thermally induced formation of worm-like, SDS-soluble, ThT-positive fibrils ^{107, 108}. While short polyQ versions of AT3 aggregate about as efficiently as the isolated Josephin domain ^{107, 108}, however, longer polyQ versions aggregate more rapidly and do so via a two-step mechanism ¹⁰⁸. In this mechanism, the initial product has similar structural features to the wormlike morphologies of the short polyQ AT3 molecules, while the final product is a more stable, amyloid-like fibril ¹⁰⁸. There are conflicting data on whether polyQ and its expansion influences the stability of the Josephin domain ^{107, 109–111}, which, if true, could influence the first step of the aggregation mechanism. The lead role played by aggregation of the Josephin domain, and/or by the intervening flexible segment ¹¹² is, however, well documented. Josephin point mutations ^{113, 114} or binding to chaperones ¹¹⁵ or normal ligands ¹¹⁴ have all been observed to suppress initial aggregation and/or amyloid formation.

Huntingtin N-terminal fragments

Huntingtin, the protein responsible for Huntington’s disease (HD), is an intracellular protein of ~ 3,200 amino acids whose polyQ sequence resides very near the N-terminus (Fig. 1). A variety of studies have implicated one or more key proteolysis events in the disease mechanism, in which release of polyQ-containing N-terminal fragments is required for the emergence of toxicity. Animals expressing expanded polyQ versions of htt N-terminal fragments like exon-1 (Fig. 1) generally exhibit aggressive HD disease features such as neurological symptoms, cellular aggregates, cellular abnormalities, and death ¹¹⁶. In vitro studies on recombinant exon-1 proteins exhibit a parallel polyQ repeat length dependent amyloid formation ⁸². In contrast with simple polyQ aggregation, however, the aggregation pathway of htt N-terminal fragments in vitro features early formation of aggregates lacking typical amyloid morphologies ¹¹⁷.

Initial investigation of htt polyQ flanking sequence effects focused on the unusual Pro-rich segment (Fig. 1). In cell models, removal of the Pro-rich sequence from exon-1 significantly enhances aggregate formation and/or toxicity ^{118, 119}. In vitro, the previously discussed ability of a C-terminal polyproline segment to reduce the aggregation rate of a polyQ sequence ²⁵ is also observed in the htt N-terminal fragment context ⁴⁸ (Fig. 5a). The polyproline effect observed in vitro can be accounted for by its ability to alter the conformational mix in the polyQ component toward less aggregation-prone conformations ^{25, 120, 121}. This conformational induction depends on details of the connection of polyproline to polyQ: a P₁₀ sequence at the polyQ C-terminus inhibits aggregation, but at the polyQ N-terminus does not ²⁵. The situation is more complicated in yeast cells, where independently expressed polyproline sequences can interfere with polyQ aggregation in trans ¹¹⁹.

Figure 5.

Kinetics of spontaneous aggregation of polyQ peptides linked to htt^NT. (a) Effect of presence and location of htt^NT; (b) effect of polyQ length on htt^NTQ_NK₂ aggregation. From reference ⁴⁸.

The mechanism of the htt^NT effect on polyQ aggregation

The addition of the 17 amino acid htt^NT sequence (Fig. 1) to the N-terminus of polyQ produces a dramatic increase in aggregation rate compared with the simple polyQ peptide ⁴⁸ (Fig. 5a). In contrast with the polyproline effect, htt^NT has the same rate enhancing effect when placed either on the N- or C-terminus of polyQ ⁴⁸ (Fig. 5a). In a series of htt^NTQ_N peptides, there remains a very strong repeat length dependence ⁴⁸ (Fig. 5b), consistent with previous observations on htt exon-1 sequences ^{82, 104}. In analogy to the AT3 system (see above), aggregation of htt^NTQ_N proteins occurs by a two-step mechanism. Rough features of this two-step mechanism are now fairly well understood (Fig. 6).

Figure 6.

Mechanism of htt^NT-mediated amyloid formation of htt N-terminal fragments. Disordered htt N-terminal fragment monomers (1; htt^NT = green; polyQ = orange; polyproline = black) reversibly form α-helix rich tetramers (2) via the htt^NT domain, and these tetramers can further reversibly assemble into octamers (3), dodecamers (4), and higher order oligomers. Any oligomer has a certain potential to undergo nucleation of amyloid structure (5), supported by the highly concentrated polyQ chains on its periphery; the propensity of any oligomer to undergo nucleation is probably linked to polyQ repeat length and the presence or absence of polyproline. Once polyQ-core amyloid nuclei are formed, amyloid elongation proceeds by monomer addition (6). As monomer concentration decreases due to fibril elongation, oligomers (2,3,4) that have not undergone nucleation dissociate to replenish the monomer pool and continue to support fibril elongation. Adapted from references ^{50, 143}.

After long incubation at relatively high concentration, htt^NT peptides without any attached polyQ form α-helix rich ^{50, 122} (Fig. 2a), pelletable ⁴⁸ aggregates. This is now visualized as a continuation of the ability of htt^NTto reversibly form α-helix rich tetramers, octomers and dodecamers as part of a native state ensemble (Fig. 2). When polyQ sequences are attached to htt^NT, pelletable oligomers tend to form more quickly in a polyQ repeat length dependent fashion ⁴⁸. Once oligomers are formed, they have the ability to stochastically rearrange to form a polyQ-core amyloid nucleus, while retaining the α-helical structure in htt^{NT 50} (Fig. 6). This enhanced polyQ amyloid formation is presumably related to the AT3 nucleation mechanism, being mediated by high local polyQ concentrations within the intermediate aggregates. This second, nucleation step appears to also be polyQ repeat length dependent ^{48, 50}. For example, an htt^NTQ₇K₂ peptide oligomer cannot undergo nucleation of polyQ amyloid structure, while an htt^NTQ₈K₂ oligomer can ⁵⁰. While nucleus formation in short polyQ htt N-terminal fragments appears to take place within large, pelletable aggregates ⁵⁰, it is possible and even likely that long polyQ htt^NTQ_N peptides undergo efficient nucleation within tetramers or octomers (Fig. 6). The critical role of α-helical bundle formation in htt^NT-mediated polyQ aggregation is further supported by inhibitor studies described below. The two-step mechanism involving an initial oligomer held in an α-helix bundle, followed by nucleation of amyloid structure in non-bundled segments, is similar in many respects to a recent proposed mechanism of amyloid nucleation in Aβ and IAPP ¹²³.

There are many details of the nucleation process to work out, but typically this event is associated with a number of ensemble observations. After nucleation, aggregation rate increases dramatically ⁴⁸, except with very low polyQ repeat lengths ⁵⁰. Aggregate morphology changes from oligomers and protofibrils to fibrils ⁴⁸ (Fig. 7). Post-nucleation aggregates (a) respond to ThT, (b) become capable of seeding elongation, (c) exhibit a less solvent exposed htt^NT sequence, (d) begin to develop β-structure in FTIR spectra, and (e) have polyQ sequences that are no longer recognizable by an anti-polyQ antibody ^{48, 50}. While most of these details were worked out with relatively short htt N-terminal fragments of the general structure htt^NTQ_NP₆K₂ and htt^NTQ_NP₁₀K₂, recent studies with chemically synthesized htt exon-1 ¹²⁴ show that these short polyproline analogs behave very similarly to htt exon-1 in aggregation rates, mechanism, and response to inhibitors (B. Sahoo, D. Singer, T. Zuchner and R. Wetzel, Ms. in preparation).

Figure 7.

Electron micrographs. (a, b) Htt N-terminal fragment aggregates isolated from tg HD Q₁₅₀ mouse brains, scale bar = 100 nm ¹⁵¹; (c) PolyQ mature aggregates produced in vitro at 37 °C from K₂Q₃₇K₂, scale bar = 50 nm ²⁰; (d) Aβ₄₀ amyloid fibrils generated at 37 °C in PBS without agitation, scale bar = 200 nm ¹⁹²; (e) aggregates of K₂Q₃₇K₂ generated in PBS solution incubated frozen at −20 °C, scale bar = 50 nm ²⁰; (f - l) aggregates of htt N-terminal fragments generated at 37 °C, as follows: (f) mature fibrils of Q₅₁ exon-1, scale bar = 100 nm ¹⁰⁴, (g) mature fibrils of Q₄₄ exon-1, scale bar = 200 nm ¹¹⁷ (h) mature fibrils of htt^NTQ₃₀P₆K₂, scale bar = 50 nm ⁴⁸; (i) initial oligomers of Q₄₄ exon-1 ¹¹⁷; (j) initial oligomers of htt^NTQ₃₀P₆K₂, scale bar = 50 nm ⁴⁸; intermediate protofibrillar aggregates of htt^NTQ₃₀P₆K₂, scale bar = 50 nm ⁴⁸; final aggregates of htt^NTQ₃₇P₁₀K₂ (S13D/S16D) ⁵⁸.

There are some alternative models for the aggregation mechanism of htt N-teminal fragments and how htt^NT contributes to it. Tam et al. posit that the role of α-helical htt^NT is to intramolecularly induce a more amyloidogenic conformation in the polyQ segment ¹²⁵. Others, based on experimental ¹²⁶ or computational ⁶¹ studies, propose that initial oligomer formation by htt N-terminal fragments is mediated primarily by direct polyQ interactions.

Implications of the htt^NT-dependent aggregation pathway

The role played by the htt^NT segment in determining the physical properties of htt N-terminal fragments may greatly affect the behavior of such peptides in cells. In addition to the possible regulatory role of htt^NT-mediated, reversible oligomer formation (see above), the role of htt^NT in mediating amyloid formation suggests that post-translational modifications within this sequence might influence the rate or extent of pathological aggregation. A potential example is in the phosphorylation of Ser13 and Ser16. A double Ser to Asp phosphomimetic mutation within htt^NT in expanded polyQ htt in a tg mouse model was found to very effectively block aggregate formation and neurodegeneration ⁵⁸. In a peptide model system, the same double Ser to Asp mutation in vitro was found to both suppress aggregation rate and to alter final aggregate morphology ⁵⁸ (Fig. 7l). These data suggest that protein phosphorylation might not always or only play a regulating role in the cell, but might sometimes lead to important cellular consequences simply due to changes in the biophysical properties of the protein substrate ^{58, 127}.

The existence of two radically different aggregation mechanisms for polyQ sequences (i.e., in simple polyQ sequences and in complex sequences like htt fragments) suggests the possibility of mixed mode aggregation under specific conditions. In fact, incubation of an htt N-terminal fragment in the presence of an inhibitor of htt^NT-mediated aggregation (below) appears to open the door to the alternative aggregation mechanism engaged by simple polyQ sequences (M. Jayaraman, A. K. Thakur, and R. Wetzel, unpublished). This may be the source of temperature-dependent aggregate conformations reported by the Tanaka group ¹²⁸. In addition to HD-specific issues, the htt^NT story has more general implications for predictions of amyloidogenic sequences, providing an example of a sequence that greatly enhances amyloid formation rates without ever engaging an amyloid or β-sheet structure ^{50, 122}.

The question of whether non-amyloid, oligomeric intermediates are on-pathway or off-pathway to amyloid formation has stimulated significant work and discussion in the field ⁹⁵, and there is almost certainly no universal answer that will apply to all amyloid assembly reactions. The assembly mechanism suggested in Figure 6 presents an interesting case, in which α-helix rich oligomers play both an on-pathway role - as the substrate within which nucleation of amyloid structure occurs ¹²⁹, and an off-pathway role – as a reservoir for the release of monomers ¹³⁰ to feed amyloid elongation as solution phase monomer is depleted ⁵⁰.

Inhibitors of Polyglutamine Aggregation

Characterization of aggregation inhibitors is important for several reasons. If accumulation of a particular type of aggregate in cells is responsible for at least some HD symptoms, then understanding inhibitor structure and mechanism of action can contribute directly to the discovery of lead compounds, and indirectly to providing a basis for inhibitor design. Even if the practical aspects of inhibitor delivery complicate an anti-aggregation approach to HD therapy, inhibitors should make excellent tools for determining aggregation mechanism and for understanding the cellular basis of aggregate toxicity. Inhibition of huntingtin aggregation is an especially interesting challenge, since, while the mechanism is quite different from that of simple polyQ amyloid formation, the polyQ core structure of the aggregates appears to be quite similar ¹²². Although high-throughput screens conducted in cell or animal models are potentially powerful ways of cutting through bioavailability issues, they also have the downside of delivering hits whose mechanisms of action are initially obscure and possibly indirect. Given the focus of this review, cell screening based inhibitors are only described if subsequent characterization indicated a direct polyQ effect.

Inhibitors of simple polyQ aggregation

Directly targeting polyQ aggregation by focusing on simple polyQ is attractive because effective inhibitors might be useful in all 10 expanded CAG repeat diseases. Burke and co-workers identified peptide based inhibitors by using phage display to select short sequences with high affinities to immobilized polyQ sequences ¹³¹. One peptide, the Trp-rich sequence QBP1, is particularly effective against polyQ aggregation in vitro ¹³¹ with an IC₅₀ in the 5 μM range ¹³². Subsequent studies characterized structure-activity relationships in this inhibitor ^{132, 133}. This peptide is effective in inhibiting aggregation and/or toxicity in cell ^{131, 134} and animal ¹³⁵ models, and is widely used to test the role of polyQ amyloid formation in the aggregation of complex proteins (see, for example reverence ⁶⁶).

The β-sheet rich structure of the amyloid end products of polyQ aggregation also has provided the basis for several structure-based designs of inhibitors. By inserting periodic Pro-Gly pairs along a polyQ sequence, it is possible to control how that sequence engages an amyloid-like structure ⁷⁴. By then including additional Pro residues at mid-β-strand locations between the Pro-Gly pairs (i.e., PGQ₉P^1,2,3, Fig. 8b), inhibitors of polyQ amyloid elongation, with IC₅₀ values in the 1–2 μM range, were obtained that are imagined to work by adding to fibril growing ends while being incapable of consolidating amyloid structure ^{87, 88, 98} (Fig. 8b). Prolines are anathema to β-sheet formation through a combination of constraining backbone torsion angles, steric clashes with β-sheet geometry, and the absence of an H-bonding N-H group ¹³⁶. The latter factor has also been used in polyQ aggregation inhibitor design by installing backbone N-methyl groups into oligoGln peptides ¹³⁷.

Figure 8.

Structure and mechanism-based inhibitors of polyglutamine amyloid formation. (a) dock and lock mechanism of amyloid fibril elongation from an encounter of the growth end of a fibril with a condensed coil polyQ monomer; (b) encounter of the growth end of the fibril with disordered PGQ₉P^1,2,3 monomer (K₂Q₉-PG-Q₄PQ₄-PG-Q₄PQ₄-PG-Q₄PQ₄K₂; proline, red; glycine, green) showing docking via the single unbroken Q₉ repeat, but an inability to rearrange into stable β-sheet in the locking step due to the mid-strand proline residues; (c) hypothetical inhibition complex when htt^NT (blue) is incubated with an htt N-terminal fragment (color scheme, see Figure 6); hypothetical inhibition complex when htt^NT-PGQ₉P^1,2,3K₂ (blue htt^NT and polyQ chain showing only the mid-strand Pro residues in red) is incubated with an htt N-terminal fragment.

Inhibitors of huntingtin N-terminal fragment aggregation

A number of library screening and structure-based design studies have been conducted with huntingtin N-terminal fragments like exon-1 as targets. Using a screening assay built around in situ proteolytic generation of htt exon-1 and a filter-trap aggregation assay ¹³⁸, Wanker and co-workers have screened several hundred thousand compounds ^{139, 140}. Two classes of inhibitors thus identified are polyphenols ^{140, 141} and benzothiazoles ¹³⁹. Polyphenols like EGCG ((−)-epigallocatechin-gallate) appear to work by rapidly directing monomeric htt exon-1 down an alternative aggregation pathway leading to stable, non-amyloid aggregates ¹⁴⁰. Two benzothiazoles have been shown to be effective, with EC₅₀ values of 40 μM, at reducing aggregate burden in organotypic brain slice cultures from the R6/2 htt exon-1 tg mouse ¹⁴². Unfortunately, at the maximum tolerated dose, one of these compounds, riluzone, is ineffective in the R6/2 mouse, in spite of reaching 100 μM levels in the brain ¹⁴².

Another series of inhibitor can be considered to be structure and mechanism based. Peptides related to the isolated htt^NT sequence prove to be excellent inhibitors of htt N-terminal fragment amyloid formation ¹⁴³. The mechanism of action appears to be via their ability to make mixed oligomers with the exon-1-like molecules ¹⁴³. Co-assembly reduces the local concentration of polyQ chains in the oligomers and thereby decreases the frequency of amyloid-like nucleus formation (Fig. 8c). Thus, sequence variants that decrease the ability to fold into an amphipathic helix decrease inhibitory power ¹⁴³. In general, these inhibitors slow nucleation but do not totally eliminate it, presumably because, statistically, there are always a few oligomers formed that are predominantly composed of htt N-terminal fragments. In an extension of this design, the molecule htt^NT-PGQ₉P^{1,2,3 –} a combination of the htt^NT segment with the elongation inhibitor PGQ₉P^1,2,3 (see above) – is an even more effective inhibitor ¹⁴³. It presumably works by co-assembling with the htt N-terminal fragments, thereby diluting their solvent exposed polyQs with an active elongation inhibitor (Fig. 8d). This mode of inhibition is similar to the proposed model for how Pro-containing variants of islet amyloid polypeptide (IAPP) inhibit WT IAPP amyloid formation ¹⁴⁴.

A related mode of inhibition involving the role of htt^NT is to disallow homo-tetramerization by complexing the htt^NT moiety of htt N-terminal fragments with another protein. There are several examples. A recent X-ray crystal structure of a complex between htt^NT and an engineered antibody fragment with anti-aggregation and anti-toxicity activity in cells shows, in a different setting, the tendency of this peptide to adopt α-helix on binding ¹⁴⁵. Molecular chaperones have been reported to reduce htt exon-1 aggregation and toxicity in mammalian cells ¹⁴⁶, and it now appears that at least some molecular chaperones do so by targeting the htt^NT sequence to suppress amyloid nucleation. TriC, for example, appears to inhibit htt exon-1 aggregation by binding and sequestering the htt^NT component ¹²⁵. It seems possible that the ability of Hsp70/Hsp40 to inhibit htt exon-1 aggregation ¹²⁶ also relies on sequestration of the htt^NT segment. This seems especially likely given the reported ability of Hsp40/Hsp70 to reduce non-amyloid oligomer formation ¹²⁶, as well as the observation that Hsp40/Hsp70 only blocks htt exon-1 aggregation when added during the lag phase ¹⁴⁷. The prospects for therapeutic exploitation of these observations are not clear. Although over-expression of Hsp70/Hsp40 suppresses neurodegeneration in a number of cell and animal studies ¹⁴⁸, mixed results are obtained in the (very aggressive) R6/2 tg mouse model ¹⁴⁸. Especially given the MoRF-like character of the htt^NT sequence ⁴⁸, there may be many other cellular proteins, besides chaperones and htt N-terminal fragments themselves, that bind to the htt^NT sequence during the normal cellular life of htt. These potentially very fluid interactions may play subtle modulating roles of htt aggregation.

Structural Features of Polyglutamine Aggregates

In vivo aggregates

As in other aggregation-associated neurodegenerative diseases, the disperse distribution and relatively light aggregate burden compared with total brain mass makes it difficult to obtain quality data on the underlying structural features of tissue-derived polyQ aggregates. PolyQ protein aggregates appear to be exclusively intracellular. The earliest observation of aggregates placed them in the neuronal nucleus ^{12, 13}. Subsequently it was recognized that aggregates are also often found outside the nucleus in the cell body ^{149, 150}.

There is limited ultra-structural data on in vivo aggregates. High resolution electron micrographs of inclusions in brain slices sometimes reveal a substructure of well-defined fibrils¹². In addition, aggregates isolated from the brains of a Huntington’s disease (HD) mouse model exhibit a range of morphologies including a number of filamentous structures similar to those observed in the in vitro assembly of huntingtin fragment aggregates ¹⁵¹ (Fig. 7a,b). At least some of the end-stage aggregates in the brain appear to have amyloid-like structures, since they are capable of specifically and strongly binding biotinylated polyQ in a recruitment stain dependent on the normally highly specific amyloid elongation reaction ¹⁵². The lack of specific stains for other types of (non-amyloid) aggregates, such as the oligomeric structures commonly observed in the early stages of the aggregation of huntingtin (htt) exon-1 and similar fragments (see below), makes it difficult to determine if these structures exist in tissue, and, if so, where they are localized and whether they persist.

Many cell models have been described in which polyQ proteins are fused to GFP. In these models, short polyQ sequences give diffuse fluorescence while long polyQs lead to one to several bright puncta either in the nucleus ¹⁵³ or cytoplasm ¹¹⁸. The limited number of inclusions per cell might be evidence for the rarity of an aggregation nucleation event ¹⁵⁴ but could also reflect the ability of the cell to organize aggregates, such as in perinuclear aggresomes ¹⁵⁵. Since the assembly of some complex polyQ proteins such as htt exon-1 proceeds with intermediate formation of non-amyloid aggregates (see below), and since different aggregated states may have different toxicities ¹⁵⁶, it is of considerable interest whether particular GFP puncta in different experiments represent accumulations of amyloid or other aggregates. AUC of cell contents ¹⁵⁷, and fluorescence methods like FRET, FRAP, FLIP, FCS and split-GFP techniques ^{51, 52, 158–160} may ultimately clarify such issues.

In vivo aggregates are often found to be associated with other cellular proteins. In some cases this might reflect the operation of cellular mechanisms directed at protein misfolding or aggregation ¹⁶¹. Due to the promiscuity of polyQ amyloid elongation ²⁰, polyQ aggregates can also effectively recruit other polyQ-containing proteins ⁸⁴. Recruitment and sequestration ¹⁶², and/or recruitment-mediated elimination ¹⁶³, of important cellular proteins such as transcription factors represent attractive mechanisms for how amyloid-like polyQ aggregates might be toxic. Such recruitment mechanisms may even sometimes contribute to identification of polyQ repeat-length dependent protein interaction partners ^{164, 165}. Although normally interpreted as evidence for specific binding partners for monomeric polyQ proteins, in at least some cases identification of binding partners may reflect the binding of proteins to already aggregated polyQ bait proteins¹⁶⁶.

Structures and properties of simple polyQ aggregates assembled in vitro

Mirroring the fine-structures of some polyQ aggregates formed in vivo, aggregates of simple polyQ peptides grown in aqueous buffers from disaggregated peptides are amyloid-like, exhibiting filamentous morphologies in EM images (Fig. 7c) ^{20, 81}. In spite of their filamentous morphologies, most aggregates grown in vitro from simple polyQ peptides do not exhibit the classical twisted fibril morphology seen in EM images, as typified, for example, by Aβ₄₀ fibrils (Fig. 7d). Other features of amyloid are well-replicated, however. Simple polyQ aggregates exhibit β-sheet features in circular dichroism ²³, FTIR ⁴⁸, X-ray diffraction ^{73, 167–169}, and solid state NMR ^{122, 170}. Although the signal is unusually low, polyQ aggregates also bind thioflavin T (ThT) and exhibit the ThT fluorescence spectral shift characteristic amyloid fibrils ⁸¹. Overall, there is no evidence that the polyQ component of simple or complex polyQ sequences in mature fibrillar aggregates exists in anything other than β-sheet rich, amyloid-like structure, in contrast to a recent proposal that α-helix-rich coiled coil might be the structural basis of such fibrils ⁴⁶.

The absence of a 10 Å reflection in the X-ray powder diffraction of polyQ aggregates led Perutz to suggest a β-helical structure for these aggregates ¹⁶⁷, and a number of computational analyses have been found to support the feasibility of such structures in polyQ sequences ^{26, 32, 91, 171–173}. Subsequent X-ray diffraction analyses, however, suggested a more conventional β-sheet structure for polyQ aggregates, although also finding, in contrast to many peptide-based amyloid fibrils ¹⁷⁴, an anti-parallel β-sheet arrangement ^{168, 169} featuring β-hairpins ¹⁶⁹. Certain features of solid state NMR ^{122, 170} and CD (J. Kardos, personal communication) spectra are also consistent with an anti-parallel β-sheet structure in polyQ aggregates. The ability of polyQ sequences interspersed at regular intervals with Pro-Gly pairs to form aggregates with kinetics, mechanistic features, and morphologies similar to unbroken polyQ has also been interpreted as indicating an anti-parallel β-sheet arrangement ⁷⁴. Several anti-parallel four stranded sheet models have been studied in silico ^{30, 31, 36, 73, 175, 176}. While there is a lot of physical evidence (see above) that such structures are not measurably populated in the monomeric state, these would appear to be reasonable models for aggregation nuclei or for the fundamental repeat unit of polyQ amyloid fibrils.

Like other amyloid fibrils, simple polyQ aggregates exhibit a characteristic concentration of monomer when the fibril formation reaction reaches equilibrium ^{45, 81}. This monomer concentration can often be approached both in the growth direction and the dissociation direction ^{45, 177} (Fig. 9), and is formally equivalent to the critical concentration, C_r, for aggregation (the concentration below which spontaneous fibril formation cannot be initiated). Even for less stable aggregates of short polyQ peptides, the dissociation of fibrils to the equilibrium position takes a very long time ^{45, 81} (Fig. 9). Since the C_r is related to the elongation equilibrium constant for fibril growth ¹⁷⁷, the C_r is an indicator of fibril stability. Not surprisingly, the C_r decreases as polyQ chain length increases, with a value of ~ 3 μM for K₂Q₂₃K₂ ⁴⁵ and 1 μM or less for longer sequences ⁸¹. This trend suggests that aggregates of long polyQ peptides are more stable than those of short polyQs, and may have implications for the thermodynamic probability of aggregate formation in the cell.

Figure 9.

Confirmation, for K₂Q₂₃K₂, of a dynamic equilibrium for amyloid fibril elongation with a characteristic end-point concentration of monomer equivalent to the C_r. From reference ⁴⁵.

There are many unresolved questions about amyloid-like polyQ aggregate structure. Although experiments with Pro-Gly interrupted polyQ sequences suggested that polyQ amyloid can be manufactured with β-sheet widths of 7–9 Gln residues ⁷⁴, it is possible that unbroken polyQ might favor significantly wider sheets ¹⁷⁰. There are several chain connectivity patterns possible in an anti-parallel β-sheet ¹⁷⁴, and it remains to be seen whether the pattern in polyQ amyloids involves β-hairpins or longer loops and whether it is uniform or variable. The structural role of the Gln sidechains has also not been established with certainty. Diffraction- or ssNMR-based models of fibrillar aggregates ^{122, 168–170, 178} or microcrystals ^{179, 180} of polyQ ^{168, 169} or Gln/Asn-rich ^178–180 peptides, as well as AFM studies of Gln-rich peptides ⁸⁶, exhibit different structural resolutions and degrees of Gln side-chain interdigitation.

Amyloid-like aggregates formed in frozen solutions

Simple polyQ peptides have the unusual ability to highly efficiently make amyloid-like aggregates in frozen solutions in PBS ^20,81. These aggregates share many properties of aggregates formed from the same peptides at 37 °C, but exhibit an atypical monofilament morphology (Fig. 7e). Perhaps because of this, compared with 37 °C aggregates, these assemblies are sometimes found to be much better seeds for elongating polyQ monomers ¹⁸¹ and are much more efficiently taken up by mammalian cells in a novel cell model for polyQ aggregate toxicity ¹⁸² and elongation ¹⁸³. Since aggregate formation depends on the amount of time incubated at −20 °C ⁸¹, this process seems to occur in the frozen solution and not as a consequence of the freezing and/or thawing processes. It is presumably brought about by the process of “freeze concentration” ^{184, 185}, whereby solutes, at freezing temperatures above their eutectic points, accumulate in liquid fissures between ice lattices resulting in high solute concentrations. The atypical monofilamentous structures of these aggregates may reflect the confined environment in which aggregation takes place or a deviant β-sheet structure within the amyloid protofilaments. These aggregates have a number of potential uses for studies of polyQ phenomena.

Non-amyloid oligomers of simple polyQ peptides

There is a lively discussion in the literature as to whether simple polyQ peptides in the class K₂Q_NK₂ are capable of forming non-amyloid aggregates, and whether such aggregates play a role in the mechanism of spontaneous amyloid growth. Some computer simulations predict the ability of polyQ sequences to make such structures ^{37, 91}, and aggregates described as non-amyloid oligomers have been observed in light scattering, EM and AFM measurements ^{77, 93}. These reports are inconsistent with other experiments, however, where oligomeric aggregates are looked for but not observed ^{23, 45}. These divergent results likely stem from differences in sample preparation. The required disaggregation protocol for synthetic polyQ peptides ^{45, 96, 186, 187} includes transient exposure to the volatile solvent hexamethylisopropanol (HFIP). This solvent is well known to dramatically modify assembly of some amyloidogenic peptides at 2–4% concentrations ^{188, 189}, and thus must be rigorously removed under vacuum as part of the disaggregation protocol. In fact, it is possible to drive the formation of homogeneous K₂Q_NK₂ oligomers in aqueous HFIP ⁹². These data are consistent with the hypothesis that non-amyloid K₂Q_NK₂ oligomers arise due to incomplete HFIP removal. This question of non-amyloid oligomer formation is an important issue for several reasons as discussed elsewhere in this review.

In vitro aggregates of complex polyQ polypeptides

Limited available data suggests that the amyloid cores of mature aggregates of simple polyQ peptides and polyQ peptides with flanking sequences are essentially identical. In ss NMR analysis, the spectral features of this core in fibrils of an htt N-terminal fragment are identical to those of a K₂Q_NK₂ peptide of similar repeat length ¹²². Analogous FTIR spectra are also very similar ⁴⁸. EM images of mature aggregates of htt N-terminal fragments ^{48, 104, 117, 151} show differences compared to aggregates of simple polyQ peptides, with filaments of greater thickness and rough edges that probably reflect the presence of coats of alternatively structured flanking sequences (Fig. 7f–h). For example, while it was expected that polyQ amyloid structure might propagate into the htt^NT flanking sequence at some point during amyloid growth, ss NMR showed the surprising result of a stable element of α-helix in a portion of the htt^NT sequence in mature fibrils ¹²². This reflects the strong tendency of this sequence to form α-helix in various associated states, even though it is disordered in the monomeric state ^{48, 50}.

In contrast with simple polyQ peptides, oligomer formation in the early stages of htt N-terminal fragment aggregation appears to be universally observed and has a fairly well-understood structural basis, as described in the section on complex polyQ peptide aggregation mechanisms. Similar small oligomers are formed from recombinant (Fig. 7i) or chemically synthesized (Fig. 7j) htt N-terminal fragments. In vitro produced protofibril-like structures (Fig. 7k) intermediate between spherical oligomers and mature fibrils resemble some aggregates isolated from tg HD mouse brains (Fig. 7b)

Conclusions

Expanded polyQ repeat diseases are often referred to as misfolding diseases, where misfolding can be meant to refer to the generation of a wide range of species, from alternative folded states of monomers to abnormal pathological self-assembled states ranging from non-amyloid oligomers to amyloid fibrils to the further self-associated inclusions seen in cell and animal models. The biophysical basis of these hypothesized states has been examined in this review.

How close are we to identifying the toxic species? Evidence for a stable or meta-stable specific folded state of expanded polyQ monomers, either in isolation or as part of a disease protein, remains elusive. At the same time, fundamental consideration of the intrinsically disordered nature of the polyQ sequence suggests that important complexes might be formed in the cell that do not depend on the existence of any populated, pre-existing, misfolded, “toxic” monomer. Thus, intrinsically disordered proteins (IDPs) are understood to be capable of highly specific coupled folding and binding ¹⁹⁰ to cellular targets whose nature would never be apparent from only a knowledge of the especially rugged folding landscape of the IDP in isolation. In the context of this hypothetical cellular target, long polyQs might make stable complexes, and short polyQs unstable complexes, simply by virtue of the amount of buried surface area, H-bonds, etc., formed in the complex. Such a model positing that “long polyQs are toxic because they are long” may be scientifically unsatisfying, but is nonetheless consistent with the intrinsically disordered character of the polyQ sequence. The definitive test of the “toxic monomer hypothesis” will be to unequivocally demonstrate molecular events and structures associated with one of more early pathological cellular consequences. In contrast to misfolded polyQ monomers, there is no shortage of demonstrated polyQ aggregates found both in vitro and in vivo. However, the same standard applies to the “toxic aggregate hypothesis” as to the toxic monomer hypothesis. Bad aggregates will have to be caught in the act; the circumstantial evidence of correlation does not provide enough evidence to lead to a conviction.

The physical chemistry of polyQ aggregation is perhaps more clear than its disease role, but is nonetheless not without controversy. PolyQ sequences are most thermodynamically stable in an aggregated state, and there are apparently a number of kinetically accessible paths, with varying efficiencies, to aggregate formation. Mechanisms for aggregation nucleation in vitro can be simple or ornate, but are ultimately driven by the high stability of the amyloid fold. Deciphering some of these mechanisms has allowed the design of structure and mechanism-based aggregation inhibitors which may allow tests of the aggregation hypothesis for expanded polyQ diseases, and ultimately perhaps an entrée into therapeutic design.

An important recent discovery is that at least one disease protein, the N-terminal fragment of huntingtin, is capable of a regular, reversible tetramerization through its htt^NT N-terminus. While this equilibrium might very well play a normal role in the cellular biology of huntingtin, it also offers a middle ground between monomers and higher-order aggregates as an expanded polyQ species which, in spite of its possible ubiquitous presence and useful functions, may take on an additional negative, and ultimately toxic, set of activities in longer polyQ molecules.

Belying the tedium of their amino acid sequences, the properties of polyglutamine segments are far from monotonous, offering continued surprises in the study of their biophysical behaviors and the search for molecular mechanisms of disease.