Location Trumps Length: Polyglutamine-Mediated Changes in Folding and Aggregation of a Host Protein

Abstract

Expanded CAG diseases are progressive neurodegenerative disorders in which specific proteins have an unusually long polyglutamine stretch. Although these proteins share no other sequence or structural homologies, they all aggregate into intracellular inclusions that are believed to be pathological. We sought to determine what impact the position and number of glutamines have on the structure and aggregation of the host protein, apomyoglobin. Variable-length polyQ tracts were inserted either into the loop between the C- and D-helices (Q_nCD) or at the N-terminus (Q_nNT). The Q_nCD mutants lost some α-helix and gained unordered and/or β-sheet in a length-dependent manner. These mutants were partially unfolded and rapidly assembled into soluble chain-like oligomers. In sharp contrast, the Q_nNT mutants largely retained wild-type tertiary structure but associated into long, fibrillar aggregates. Control proteins with glycine-serine repeats (GS₈CD and GS₈NT) were produced. GS₈CD exhibited similar structural perturbations and aggregation characteristics to an analogously sized Q₁₆CD, indicating that the observed effects are independent of amino acid composition. In contrast to Q₁₆NT, GS₈NT did not form fibrillar aggregates. Thus, soluble oligomers are produced through structural perturbation and do not require polyQ, whereas classic fibrils arise from specific polyQ intermolecular interactions in the absence of misfolding.

Introduction

The expanded CAG repeat diseases consist of nine neurodegenerative disorders that are characterized by progressive loss of motor function and eventually cognitive decline, the most common of which is Huntington's disease (1). In affected individuals, a disease-specific protein is expressed with an abnormally long polyglutamine (polyQ) domain. The proteins share no other sequence or structural homology (2). Nuclear and/or cytoplasmic aggregates of the polyQ protein and/or its proteolytic fragments are invariably present in the disease state (3). These protein aggregates are hypothesized to cause neurodegeneration, although the mechanism for toxicity remains elusive. Two leading theories suggest that sequestration of transcription factors into aggregates (4,5) or overload of the ubiquitin-proteosomal system (6) may play a role.

The length of the expanded polyQ domain is inversely correlated with the age of onset of symptoms. The correlation is striking enough that it is common to describe a threshold length for disease, which varies from ∼20 to ∼60 glutamines depending on the specific disorder (1). Both the length of the polyQ domain (7–10) and its context within the protein (11–14) influence the propensity for aggregation. Some investigators have argued that destabilization of the native protein structure by expanded polyQ drives aggregation (15–17). Others have reported that increasing the polyQ length does not further destabilize the protein but still enhances aggregation (18), which would argue for the dominant role of polyQ-specific intermolecular association. Simulations suggest that the folding/unfolding transition broadens with polyQ length, increasing the lifetime of partially folded, aggregation-prone intermediates (19).

We set out to conduct a systematic study of the relative importance of the length of the polyQ domain, and the context into which the domain is placed, in terms of misfolding and aggregation. The ideal host protein for such biophysical studies is readily expressed, purified, and folded, and has well-characterized structural features. Unfortunately, in general, the proteins that are associated with expanded CAG diseases do not meet these criteria. With a few exceptions, little is known about their native structure or folding pathways, and many are difficult to express recombinantly as full-length proteins or without fusing to a solubilizing protein (2,3). We therefore chose to conduct our studies using a host protein that is not associated with disease. Precedence for this approach was established by previous studies that used glutathione S-transferase (20–22), thioredoxin (23,24), CRABP1 (8,17), Staphylococcal protein A (14), and Sup35 (25). This approach is also justified biologically by experiments wherein symptoms of classic polyQ neurodegenerative disease were induced by incorporation of expanded polyQ domains into an otherwise innocuous protein (26).

We chose to use apomyoglobin as our host protein for several reasons. First, apomyoglobin is readily expressed in Escherichia coli, and a wealth of information on its solution structure, stability, and folding pathway is available (27–30). Second, the native fold contains α-helices and a significant amount of disorder, reflecting the major secondary structural elements of the CAG-disease-related proteins (31–34). Third, we speculated that a marginally stable protein like apomyoglobin (28) would serve as a better model of amyloidogenic polyQ-containing proteins because the disease-related proteins are relatively unstable (15,35) and may populate multiple folds (36). Finally, others have used myoglobin as a model for amyloidogenic proteins (37–39) and as a host protein for polyQ (9,40,41).

We generated a small library of apomyoglobin mutants with variable polyQ length and position. Two positions were selected: the flexible surface loop between the short C and D helices (Q_nCD, where n = number of glutamines), and the N-terminus (Q_nNT). The mutants contained eight to 46 glutamines (Fig. 1). Unique to our work (to our knowledge), we created control mutants containing a glycine-serine repeat (GS₈) in place of polyQ, to assist in decoupling contributions of structural perturbation to aggregation from specific polyQ-driven interactions. We then probed changes in secondary and tertiary structure, as well as the aggregation kinetics and aggregate morphology.

Figure 1.

Myoglobin crystal structure and mutation sites (PDB ID 1MBN). In the apo form, the F-helix is unstructured. For Q_nCD, Q_n was inserted between D45 and K48, deleting R46 and F47. For Q_nNT, Q_n was inserted between M1 and V2.

Materials and Methods

Protein production and purification

We generated most of the Q_nCD mutants using a previously described procedure (42). For Q₈CD, GS₈CD, Q₈NT, Q₁₆NT, Q₃₂NT, and GS₈NT, recombinant proteins were expressed as described in the Supporting Material. Protein was recovered from the cell lysate supernatant and purified by nickel affinity chromatography. The proteins were refolded by rapid dilution, concentrated by ultrafiltration, and further purified by size exclusion chromatography (SEC; see Supporting Material). Unless otherwise specified, freshly concentrated protein samples in citrate buffer were immediately filtered with a 0.45 μM polyvinylidene fluoride filter, and the concentration was adjusted to ∼6–10 μM.

Circular dichroism

Freshly concentrated protein samples were prepared in 10 mM sodium phosphate, pH 6.8. Circular dichroism (CD) spectra were collected on an Aviv 202SF CD spectrometer at 20°C. The spectra were fit to a reference set (SP37), and secondary structure was calculated using the CONTINLL algorithm in CDPRO.

Fluorescence spectroscopy

Tryptophan fluorescence emission spectra were obtained with a QuantaMaster Series spectrofluorometer (PTI, Birmingham, NJ), with excitation at 295 nm. The center of spectral mass (λ_CSM) was calculated as described previously (44). A stock solution of 8-anilino-1-naphthalene sulfonic acid (ANS, 800 μM; AnaSpec, San Jose, CA) was diluted 10-fold with sample to yield an ANS/protein ratio of 8:1. Fluorescence spectra were obtained with excitation at 350 nm, and ANS λ_CSM was calculated using emission spectra from 410 to 500 nm. Unfolding in urea was followed by Trp fluorescence as described in the Supporting Material.

Monomer loss and sedimentation kinetics

Freshly prepared samples were injected onto a Superdex 75 PC 3.2/30 size exclusion column (GE Healthcare, Piscataway, NJ). Detection was made by absorbance at 280 nm and peak area was obtained by integration; the mean error was 4.5%. Samples were aliquoted into low-retention microcentrifuge tubes (Fisher, Pittsburgh, PA), sealed with paraffin film, and incubated at 37°C. At regular intervals, a sample tube was withdrawn and centrifuged at 19,500 rcf for 30 min. The supernatant (top 75%) was removed and subjected to SEC analysis. To determine the fraction of soluble protein, we determined the mass concentration in the supernatant in triplicate with a bicinchoninic acid (Pierce, Rockford IL) assay, using a BioTek (Winooski, VT) EL 800 Universal Microplate Reader.

Transmission electron microscopy

For transmission electron microscopy (TEM), samples were incubated at 37°C for ∼3 weeks. A drop was applied to a pioloform-coated grid, stained with methylamine tungstate stain, and then imaged with a Philips CM120 scanning transmission electron microscope (Philips, FEI, Hillsboro, OR).

Results and Discussion

Protein production, refolding, and purification

Qualitatively, the yield of protein was lower for the Q_nCD mutants than for the wild-type (WT), and yields decreased noticeably as the length of the polyQ domain increased. In contrast, yields of Q_nNT were at least as good as that of WT, and there was no drop-off in yield with polyQ length. WT spontaneously refolded to monomers after rapid dilution from denaturing buffer, but a small peak eluting at higher molecular weight was detected (Fig. S1), consistent with published results and attributed to nonnative aggregation during refolding (38). Similar peaks were detected for all Q_nCD mutants, with the fraction increasing significantly with increasing polyQ length. No high-molecular-weight peaks were observed for any Q_nNT mutants. Others have reported considerable difficulties in expressing folded and soluble polyQ-containing proteins (14,31,45). Our data illustrate that expression and refolding are strongly dependent on polyQ context. Positioning the domain in the CD loop significantly reduced both protein expression and efficient refolding in a length-dependent manner, whereas placing the domain N-terminally enhanced efficient refolding.

The monomer fraction was isolated and concentrated before further characterization. Most preparations remained fully monomeric after concentration, although for a few Q_nCD mutants we observed small oligomer peaks even after SEC purification. Most preparations were at least ∼95% pure by sodium dodecyl sulfate polyacrylamide gel electrophoresis (not shown). For the longest Q_nCD mutants, we observed some protein contaminants after SEC purification and concentration, identified by mass spectrometry as E. coli ribosomal proteins. The purity of these contaminated samples is estimated at ∼85%. The residual heme content was assayed as 2% in WT and ∼0.5% in all of the Q_nCD mutants. The heme content of Q_nNT was a weak function of length, with 0.7% in Q₈NT, 1% in Q₁₆NT, and 2.5% in Q₃₂NT.

Tanaka et al. (9,40,41) created myoglobin mutants with polyQ sequences of various lengths inserted into the CD loop. Their system differs from our Q_nCD mutants in several ways: 1), their insert included 16 residues from ataxin-3; 2), they did not specifically report removal of oligomers that could have formed during purification (9); and 3), their proteins contained variable amounts of heme (9,41). We took care to ensure that little to no heme remained, because holomyoglobin is more stable and considerably more resistant to aggregation than is apomyoglobin (46), and any variability in heme content would have complicated interpretation of our results.

Effects of polyQ location and length on secondary structure

CD spectra were collected for WT and all mutants (Fig. S2). The WT spectra exhibited minima at 209 nm and 222 nm, indicative of a predominantly α-helical protein. There was slight deepening of these minima with Q_nNT. In contrast, with Q_nCD, there was a length-dependent flattening of these minima as well as a decrease in the maximum at 194 nm. We estimated the distribution of secondary structural elements by quantitative analysis of CD spectra. The fitted values for WT agreed well with solution structures obtained from NMR (28,29). We calculated the number of residues in each secondary structure element by multiplying the fraction of that element by the total number of residues in that mutant, and then calculated the change in the number of residues adopting each secondary structure relative to WT (Fig. 2).

Figure 2.

Change in the number of residues in secondary structural elements, relative to WT. The fraction of each secondary structure element was determined as described, and then multiplied by the total number of residues in each protein to calculate the number of residues of that element in the protein. This number was compared with the calculated average number of WT residues (n = 5) in that secondary structural element. Error bars represent the maximum and minimum of two independent measurements, with the exception of Q₁₆CD (95% confidence interval) and Q₃₄CD (three independent measurements).

With Q_nNT, there was a modest increase in the calculated number of helical residues compared with WT, but little change in β-sheet or disorder. These data suggest that the added glutamine residues may adopt a partially α-helical conformation when placed at the N-terminus. Alternatively, it is possible that the addition of polyQ at the N-terminus increases the helical content of the protein core, but this seems less likely.

Significantly different effects were obtained with Q_nCD, and these effects were length-dependent. For Q₈CD, there was no change in α-helix or β-sheet, and the increase in the number of disordered residues was the same as the increase in the total number of residues. The simplest interpretation is that the native secondary structure of apomyoglobin is not disturbed and the Q₈ sequence is disordered. With intermediate-length mutants (n = 16–28), there was a loss of α-helix, a slight increase in β-sheet, and a larger increase in disordered residues. We hypothesize that insertion of Q₁₆ to Q₂₈ in the CD loop results in unwinding of the short adjacent helices (four and five residues, respectively). As the polyQ length increased further (n = 34–46), the loss of helical residues remained about the same, but there was a larger increase in the number of β-sheet residues. A reasonable interpretation of these data is that the polyQ insert along with the flanking residues adopts some β-sheet character as the insert becomes longer. These experiments do not allow us to determine whether the additional β-sheet content is restricted to the erstwhile helices of the host protein or to the inserted polyQ sequence. Overall, there was not a sharp demarcation but rather a gradual transition in secondary structure between shorter and longer polyQ domains.

Our results can be usefully compared with literature reports. Bevivino and Loll (15) reported a loss in α-helix and increase in random coil with an increase in Q length in ataxin-3. PolyQ fused to glutathione-S-transferase was flexible and disordered independent of length (20,22). Similarly, polyQ fused to the C-terminus of Staphylococcal aureus protein A (SpA) B domain was disordered independent of length, and there was no effect on the conformation of the host protein (14). Nagai et al. (23) fused polyQ to the C-terminus of thioredoxin, a β-sheet protein, and observed increased helical content in a Q-length-dependent manner. In the huntingtin exon 1 fragment, Q₁₇ adopted a heterogeneous mix of conformations, including α-helix, extended loop, and disordered (47). Modest increases in helical content due to polyQ have been reported for an androgen receptor fragment (48) and myoglobin mutants (41). These results illustrate that polyQ conformation is strongly context-dependent, and polyQ is almost universally either disordered or α-helical.

Effects of polyQ location and length on tertiary fold and stability

Apomyoglobin contains two tryptophans, W7 and W14, both of which are within the N-terminal helix. W14 contributes most to the fluorescence emission intensity because W7 fluorescence is quenched in the natively folded protein. W7 is partially solvent-exposed, and W14 is in a predominantly hydrophobic environment (49).

For all Q_nNT mutants, Trp λ_CSM was blue-shifted slightly relative to WT (Fig. 3 A), suggesting partial burial of W7 into a more hydrophobic environment. In contrast, for Q_nCD, λ_CSM was red-shifted, indicative of increased exposure to a polar environment and partial loss of native tertiary structure (41). For both Q_nCD and Q_nNT, emission intensity was Q-length-dependent (Fig. 3 B). Emission intensity was greatest for the mutants with the shortest polyQ length, and then decreased as the Q length increased. The emission intensity of Q₄₆CD was well below that of WT. One possible explanation for the length-dependent loss of Trp emission intensity is increased quenching with glutamine (50), as this loop becomes long enough to interact with the N-terminal helix.

Figure 3.

Trp fluorescence spectroscopy. All data points are the average of two experiments, with the exception of Q₂₀CD, Q₂₄CD, Q₃₄CD, and Q₃₈CD, for which three independent experiments were performed. Error bars represent the maximum and minimum values measured. (A) Center-of-spectral-mass wavelength, calculated using emissions from 320 to 360 nm. (B) Fluorescence emission intensity at maximum wavelength.

To further assist in interpreting these data, we partially unfolded WT in urea. At 2.5 M urea (20% unfolded), WT exhibited a red shift and an increase in Trp emission intensity (not shown) that was very similar to that observed for Q₈CD, Q₁₆CD, and Q₂₀CD in the absence of urea. This suggests that the polyQ domain inserted into the CD loop causes partial (∼20%) unfolding of apomyoglobin. If WT is unfolded further by increasing the urea concentration, Trp λ_CSM and intensity both increase further (not shown). This is not what we observed with increasing Q length, suggesting that although the polyQ insert triggers partial loss of tertiary structure, lengthening the polyQ domain does not push the protein toward more unfolding.

Unfolding of the mutants during urea denaturation was studied by monitoring Trp fluorescence. Q_nNT mutants were slightly more resistant to unfolding than WT as measured by the increased concentration of urea (+0.4M) required for 50% unfolding (C_m). In contrast, Q_nCD mutants were less resistant to unfolding than WT (C_m decreased by 0.6 M). There was little or no polyQ length-dependent effect on C_m (Fig. S3). The unfolding curve was broader for Q_nCD than for WT or Q_nNT mutants (Fig. S4). This is suggestive of less sharply defined transitions for Q_nCD along the unfolding pathway and the presence of a conformational ensemble rather than a single native fold, consistent with simulations (19).

ANS binds strongly to the hydrophobic heme pocket of apomyoglobin (51), and its fluorescence can be used to confirm the existence of the pocket, which forms only upon attainment of the correct tertiary fold. ANS also detects hydrophobic patches that are exposed when a protein partially unfolds. ANS emission maximum at 455 nm and strong fluorescence intensity confirmed formation of the heme pocket for WT (not shown). All of the Q_nNT mutants retained the ANS spectra of WT, indicating that addition of polyQ to the N-terminus did not interfere with formation of the pocket (Fig. 4). In contrast, the ANS fluorescence wavelength was markedly red-shifted for all Q_nCD independently of Q length (Fig. 4 A), and the emission intensity was measurably decreased in a weakly length-dependent manner (Fig. 4 B). This result demonstrates loss of the heme pocket in Q_nCD. The decrease in ANS fluorescence with increasing Q length suggests that the hydrophobic patches exposed by partial unfolding are lost, either because of further unfolding or because these patches are covered by the long polyQ segment, as suggested by simulations (52). The changes in ANS fluorescence upon introduction of residues to the CD loop are very similar to those seen in WT upon partial unfolding with 2.5 M urea (not shown). We conclude that Q_nCD suffers partial loss of tertiary structure relative to WT, whereas native tertiary structure is retained in Q_nNT.

Figure 4.

ANS fluorescence spectroscopy. All data points represent the average of two experiments, with the exception of Q₃₄CD, for which three independent measurements were collected. Error bars represent the maximum and minimum values measured. (A) Center-of-spectral-mass wavelength, calculated using emission spectra from 410 to 500 nm. (B) Fluorescence emission intensity at maximum wavelength.

Several other groups observed changes in tertiary structure or stability with polyQ proteins in previous studies, and most reported a decrease in stability. Robertson et al. (14) observed a large decrease in stability, in a Q-length-independent manner, when the polyQ domain was placed in a loop between two helices in the SpA B domain. However, they were unable to produce proteins with an interior sequence longer than 25 glutamines, and thus (in contrast to our study) did not probe the aggregation properties of these proteins. A partial loss of tertiary structure and/or a decrease in chemical or thermal stability with polyQ-containing proteins was also observed in studies involving CRABP I-httexon1 fusion proteins (8) and ataxin-3 (15,18). Several studies reported no effect of polyQ on stability (14,22). We are aware of only one study in which polyQ increased stability (measured as an increase in the temperature required to induce aggregation), with C-terminal attachment to glutathione S-transferase (21). We conclude that polyQ is more readily accommodated by a host protein if it is placed on the terminus rather than in a spatially restricted area, such as an interior loop.

Non-polyQ-specific changes in secondary and tertiary structure

To test whether any sequence inserted into the CD loop would cause a structural perturbation, we produced a mutant containing a GlySer octet repeat, GS₈CD. We chose to use glycine and serine because they are flexible, hydrophilic residues that are frequently found in disordered loops at protein surfaces (53). We analyzed CD spectra, Trp and ANS fluorescence spectra, and urea denaturation curves for GS₈CD and compared the data with those obtained from the two mutants with inserts of identical length, Q₁₆CD and Q₁₆NT (Table 1). Changes in secondary and tertiary fold for GS₈CD and Q₁₆CD relative to WT were very similar, and quite distinct from the (smaller) changes with Q₁₆NT. These data demonstrate that the partial loss of structure observed in the Q_nCD mutants does not specifically require polyQ.

Table 1.

Secondary and tertiary structure changes with 16-mer insert, relative to WT

	Change in number of residues			Change in λCSM (nm)		Change in [Urea] (M)
	α-helix	β-sheet	coil/turn	Trp	ANS	at 50% unfolding
Q16-NT	+7	+4	+6	−0.25	+0.8	+0.5
Q16-CD	−13	+10	+18	+0.49	+6.2	−0.6
GS8-CD	−8	+7	+15	+0.34	+5.6	−0.4

Monomer loss and sedimentation

The WT, three Q_nCD mutants (n = 8, 16, or 38), three Q_nNT mutants (n = 8, 16, or 32), and controls GS₈CD and GS₈NT were selected for aggregation studies. Samples were monitored for monomer loss and soluble oligomer formation by SEC, and for formation of insoluble aggregates by sedimentation.

Representative chromatograms are shown in Fig. 5. No monomer loss was observed for WT over the 4-week test period (Fig. 5 A). In contrast, for Q_nCD, rapid monomer loss was indicated by a decrease in the monomer peak, which coincided with the appearance of an oligomeric peak eluting at the void volume (70 kDa). For Q₁₆CD (Fig. 5 B), the monomer peak gradually disappeared while the oligomer peak gradually increased, whereas for Q₈CD (Fig. 5 D), the monomer peak disappeared and the oligomer peak appeared within 24 h, followed by a slow loss of oligomer. For all Q_nCD mutants, the monomer loss kinetics was biphasic, with a rapid initial loss followed by a slower decline (Fig. 6 A). Surprisingly, monomer loss did not monotonically increase with Q length. The greatest monomer loss was observed for the shortest mutant, Q₈CD.

Figure 5.

Size-exclusion chromatograms as a function of time. Curves from bottom to top represent chromatographs for injections at 0, 2, 7, 14, 21, and 28 days. (A) WT. (B) Q₁₆CD; no data at 28 days. (C) Q₈NT chromatographs (inset: magnification of the soluble oligomer peak); no data at zero days. (D) Q₈CD; no data at 21 days.

Figure 6.

Aggregation kinetics. Samples were simultaneously analyzed for monomer loss and sedimentation: Q₈CD (▪), Q₁₆CD (▴),Q₃₈CD (●), Q₈NT (□), Q₁₆NT (▵), Q₃₂NT (○), GS₈CD (×), and GS₈NT (+). Solid lines: Q_n-CD and Q_n-NT; dashed lines: GS₈ controls. (A) Monomer loss as a fraction of the initial monomer peak area. Initial monomer peak areas are an average of peak areas calculated from three injections. Peak areas for days 1–5 were calculated from single injections. Peak areas for days 7, 14, 21, and 28 are averages of two injections. Error bars from multiple injections were insignificant (<1.5%) and are not shown. (B) Sedimentation kinetics data. All data points are averages of three measurements. Error bars were small and are not shown for clarity.

The loss of soluble protein, as measured by sedimentation, lagged considerably behind the loss of monomer as detected by SEC (Fig. 6 B). Consistent with the monomer-loss data, the greatest sedimentation was observed with Q₈CD. After 4 weeks, only 40% of Q₈CD remained soluble, compared with 80% of Q₃₈CD and 90% or more for all other mutants. The difference between the monomer-loss and sedimentation kinetics can be attributed to the formation of soluble aggregates.

Partially unfolded WT apomyoglobin is prone to aggregation (38); therefore, we propose that aggregation of Q_nCD is directly linked to partial loss of native fold. With Q₈CD there was no loss of native secondary structure, but there was partial loss of tertiary structure and surprisingly aggressive aggregation. We hypothesize that in Q₈CD, rapid aggregation is driven by exposure of hydrophobic patches due to partial loss of tertiary fold. With Q₁₆CD there was a net loss of native helix, and rapid initial association into oligomers, but little sedimentation. The lower sedimentation with Q₁₆CD compared with Q₈CD could be due to the loss of helix. It is possible that the unwinding of the C- and D-helices provides sufficient additional entropy to maintain the oligomers in a soluble state. These data are usefully compared with literature reports of greater unfolding and decreased aggregation of apomyoglobin with increasing temperature (37,54). Thus, inserting Q₈ or Q₁₆ into the CD loop appears to be functionally equivalent to increasing the temperature. As the Q length increased further to 38, the loss of native helix was accompanied by increased nonnative refolding to β-sheet. We hypothesize that this refolding may explain why sedimentation is detected with Q₃₈CD but not with Q₁₆CD. The monomer-loss and sedimentation kinetics for the control mutant, GS₈CD, was very similar to that of its Q16 cousin (Fig. 6). This result demonstrates that partial unfolding and aggregation in the CD mutants is driven by the structural perturbation brought about by the insertion, and does not require specific intermolecular polyQ association.

For Q_nNT, the monomer peak area was largely constant over time (Fig. 5 C). However, close examination revealed the existence of a very small oligomer peak at the void that appeared only after several days (Fig. 5 C, inset). In contrast to the rapid initial loss of monomer with the Q_nCD mutants, there was an apparent lag period before the onset of modest monomer loss for Q_nNT (Fig. 6 A). Despite the retention of secondary and tertiary structure, the Q_nNT mutants experienced some monomer loss, whereas the WT did not. Very little sedimentation was measured (Fig. 6 B).

Aggregate morphology

No aggregates were observed in TEM images of freshly prepared WT or mutants, or for aged WT, across several fields of view (not shown). With Q_nCD aged for 3 weeks at 37°C, we observed numerous small, flexible, chainlike aggregates, 100–200 nm in length, as well as some annular aggregates (Fig. 7, A, B, and D). These Q_nCD aggregates have a fuzzy, beads-on-a-string appearance, and such aggregates are frequently termed protofibrils. Of interest, these aggregates are morphologically very similar in length, diameter, and flexibility to those obtained by thermal denaturation of apomyoglobin (54). For Q₈CD and Q₃₈CD, oligomers eventually coalesced to form larger, amorphous aggregates (not shown), which likely accounts for the material lost by sedimentation (Fig. 6 B). Aggregates of GS₈CD closely resembled those of the Q_nCD proteins (Fig. 7 C). This follows the pattern of all other assays in which the GS₈CD control exhibited properties very similar to those of Q₁₆CD.

Figure 7.

TEM images taken after incubation at 37°C for ∼3 weeks, except for GS₈NT, which was incubated for ∼2 months. All images are representative of several fields of view. Reference bars are 200 nm. (A) Q₈CD. (B) Q₁₆CD. (C) GS₈CD. (D) Q₃₈CD. (E) Q₈NT. (F) Q₁₆NT. (G) Q₃₂NT (inset: Q₃₂NT filamentous network). (H) GS₈NT.

With aged Q_nNT, we observed aggregates of starkly different morphology. With Q₈NT and Q₁₆NT, we saw highly ordered rigid fibrils, several microns in length and ∼25 nm in width, that appeared to be bundles of aligned filaments with a twisted morphology (Fig. 7, E and F). The fibrils were sparse, and we had to examine many fields of view to detect them. Q₃₂NT aggregates were much more abundant than Q₈NT and Q₁₆NT aggregates, and tended to be somewhat shorter (Fig. 7 G). These aggregates were frequently assembled into a filamentous network (Fig. 7 G, inset) very similar in appearance to that formed from TBP-Q₃₈ (55). Together with the observation that no oligomers were apparent in Q_nNT until after a lag period of several days (Figs. 5 C and 6 C), these results suggest that Q_nNT, unlike Q_nCD, self-associates via a classic nucleation-elongation pathway. We hypothesize that the ratio of nucleation/elongation rate constants increases with increasing polyQ length: slow (rare) nucleation leads to fewer, longer aggregates (as in Q₈NT or Q₁₆NT), whereas faster (more frequent) nucleation results in more, shorter aggregates (as in Q₃₂NT). We speculate that the lower number density of Q₈NT and Q₁₆NT facilitates diffusion and alignment of filaments into fibrillar bundles in an orderly fashion, whereas the more numerous Q₃₂NT filaments become kinetically trapped in a loose network because of crowding. We also examined the morphology of aged GS₈NT. Strikingly absent from these TEM images is the presence of any fibrils. However, some short, semiflexible aggregates, similar in size and shape to those of the GS₈CD and Q_nCD aggregates, can be observed (Fig. 7 H).

We hypothesize that the morphological differences between Q_nCD and Q_nNT aggregates arise from differences in the driving force for aggregation. Specifically, protofibrils arise from rapid association of partially misfolded protein via exposed hydrophobic patches. Because we saw virtually identical structural perturbations, aggregation kinetics, and aggregate morphology with Q₁₆CD and GS₈CD, we conclude that oligomerization is independent of the amino acid composition of the insert. In contrast, Q_nNT fibrils arise not from partial unfolding or misfolding but from specific polyQ-mediated intermolecular interactions. This conclusion is supported by the observation that the Q_nNT mutants were at least as stable as the WT, which does not aggregate, and GS₈NT did not form fibrils.

Our data suggest that nonspecific partial loss of native structure leads to protofibrils, whereas polyQ-mediated association produces classic fibrils. There is some support in the literature for this notion. Some groups have reported that ataxin-3 with expanded polyQ is destabilized and quickly aggregates into chains similar in appearance and dimensions to those we observed for Q_nCD (15,56,57). Other researchers observed no destabilization of ataxin-3 with polyQ expansion, and aggregation into long fibrils of 20–45 nm diameter, similar in appearance to our Q_nNT fibrils (18). Ignatova and Gierasch (8) and Ignatova et al. (17) observed a shift in the aggregate morphology of a CRABP I-httexon1 chimera over time from spherical to curvilinear to fibrillar. They suggested that aggregation is initially driven by the aggregation-prone CRABP I domain but then later is dominated by the polyQ domain. This conclusion is similar to those reached by Bulone et al. (21) in their study of GST-polyQ fusion proteins, and by Ellisdon et al. (16,57). Both chain-like oligomers and ordered fibrils have been observed with huntingtin exon 1 fragments (58–60), with the more tightly packed, fibrillar aggregates dominating as the length of the polyQ domain increased (59,60).

Conclusions

We conducted a systematic investigation of folding and aggregation in polyQ-containing proteins. We produced a library of proteins with polyQ lengths ranging from sub- to suprapathological, and with inserts at two positions: the N-terminus and an interior loop. Of importance, we included two control mutants so that we could differentiate between the generic effects of insertion and specific polyQ effects. To our knowledge, no other studies of polyQ-mediated protein misfolding and aggregation have included these kinds of controls.

The location of the polyQ domain had a dominant influence on the nature of the effect of the insert on protein folding and aggregation, whereas the length of the domain modulated the extent of these effects. One of our most striking findings is the difference in morphology between aggregates made from Q_nCD versus Q_nNT. Our data suggest that, when placed in an interior loop, polyQ destabilizes native structure and accelerates aggregation primarily through partially folded intermediates and not through specific polyQ-polyQ association. In contrast, polyQ placed at the N-terminal site does not disturb the native structure but drives assembly into ordered fibrils through specific polyQ-mediated association.

We propose that there are two modes of aggregation of polyQ-containing proteins. One mode, epitomized by the Q_nCD mutants, is driven by hydrophobic interaction caused by destabilization and partial loss of native fold, leading to protofibrils and other oligomers. The initial aggregation kinetics is rapid, and, of importance, aggregation does not specifically require glutamines. In the other mode, epitomized by the Q_nNT mutants, association into filaments and fibrils follows classic nucleation-elongation kinetics and requires specific polyQ intermolecular interactions. We speculate that the second mode is operable only when polyQ is in a less-restricted location, e.g., at the terminus or in a highly flexible segment. In this light, it is interesting to note the strong bias toward placement of repeat amino acid motifs in disease-related proteins near the N- or C-terminus (61,62) and/or in highly disordered segments (34). There is growing evidence that aggregate morphology directs toxicity, although the specific morphology required for toxicity remains a matter of debate (23,59,63). Thanks to their distinct and controllable conformational and aggregation properties, the mutants we created may prove to be useful tools for teasing out the morphology-toxicity linkage.