Kinetic partitioning of protein folding and aggregation - Nature

articles

© 2002 Nature Publishing Group http://structbio.nature.com

Kinetic partitioning of protein folding and aggregation Fabrizio Chiti1,2, Niccolò Taddei1, Fabiana Baroni1, Cristina Capanni1, Massimo Stefani1, Giampietro Ramponi1 and Christopher M. Dobson2,3 Published online: 22 January 2002, DOI: 10.1038/nsb752

We have systematically studied the effects of 40 single point mutations on the conversion of the denatured form of the / protein acylphosphatase (AcP) into insoluble aggregates. All the mutations that significantly perturb the rate of aggregation are located in two regions of the protein sequence, residues 16–31 and 87–98, each of which has a relatively high hydrophobicity and propensity to form -sheet structure. The measured changes in aggregation rate upon mutation correlate with changes in the hydrophobicity and -sheet propensity of the regions of the protein in which the mutations are located. The two regions of the protein sequence that determine the aggregation rate are distinct from those parts of the sequence that determine the rate of protein folding. Dissection of the protein into six peptides corresponding to different regions of the sequence indicates that the kinetic partitioning between aggregation and folding can be attributed to the intrinsic conformational preferences of the denatured polypeptide chain.

Polypeptides that fold into the compact globular structures of natural proteins contain a large proportion of hydrophobic residues in their sequences. Although these residues in the fully folded forms are largely buried in the protein core, their presence renders such proteins highly prone to aggregation when unfolded or partially folded1. Despite the fact that the ability of macromolecules to self-associate in a controlled manner to generate functional complexes is an essential feature of all biological systems, the avoidance of unproductive aggregation has undoubtedly been a dominant feature in the evolution of life1–3. This problem is particularly significant for newly synthesized proteins before they have the opportunity to fold to their native structures. Indeed, a typical cell uses a large fraction of its metabolic energy in combating the unwanted self-association of incompletely folded proteins, particularly through the actions of molecular chaperones, many of which are ATP dependent4. Moreover, a large number of human disorders, ranging from type II diabetes to Parkinson’s and Alzheimer’s diseases, are associated with protein aggregation resulting from aberrant folding or processing events5–7. In these cases, aggregate formation may result not only in the absence of correctly functional proteins but also in cellular toxicity caused by the aggregates themselves8,9. Despite its fundamental biological importance, little is known about the molecular basis or specificity of protein aggregation. Here we describe a strategy to investigate the relative importance of different regions of a natural protein sequence in the process of aggregation. Given the complexity of the problem, we have used human muscle acylphosphatase (AcP), a relatively simple α/β protein consisting of 99 residues and lacking complicating factors, such as disulfide bridges or bound cofactors10. Of particular importance is the observation that this protein aggregates in a very well-defined manner under appropriate conditions, ultimately forming highly organized amyloid fibrils of the type asso-

ciated with protein deposition diseases11,12. Our strategy is based on determining the relative aggregation rates of mutational variants of AcP in which conservative single amino acid substitutions are made throughout the whole sequence. Therefore, it is similar in concept to the protein engineering method of investigating protein folding13, which has transformed our detailed understanding of this process14–16. As a complementary approach, we have also dissected the sequence of AcP by synthesizing six peptides that correspond to different regions of the protein and investigating their conformational and aggregational properties. Strategy designed to probe the aggregation of AcP We initially studied 34 mutants of AcP (legend to Fig. 2), allowing the polypeptide chain to be probed approximately every three residues along its sequence. Because the globular fold generally protects a protein against self-association by burying the aggregation-prone main chain and hydrophobic residues in its close-packed interior, amino acid replacements often favor aggregation by destabilizing the native state, allowing unfolded or partially unfolded species to be significantly populated12,17–19. For this reason we probed the rate of aggregation of the polypeptide chain in aqueous solutions containing 25% (v/v) trifluoroethanol (TFE), which unfolds even the most stable AcP mutants studied here but still allows aggregation to occur11,12. This procedure permits any changes in aggregation rate upon mutation to be attributed entirely to the effects of the amino acid substitutions on the intrinsic aggregation properties of the ensemble of denatured conformations. We have previously shown that aggregates formed from wild type AcP in the presence of TFE and from highly destabilized mutants in its absence are essentially identical in their underlying structures12. Moreover, the mechanism of protein folding in the presence of TFE is thought to be very closely similar to that

1Dipartimento di Scienze Biochimiche, Università degli Studi di Firenze, Viale Morgagni 50, 50134 Firenze, Italy. 2Oxford Centre for Molecular Sciences, University of Oxford, New Chemistry Laboratory, South Parks Road, Oxford OX1 3QT, UK. 3Present address: Department of Chemistry, University of Cambridge, Lensfield Road, Cambridge, CB2 1EW, UK.

Correspondence should be addressed to C.M.D. email: [email protected] nature structural biology • volume 9 number 2 • february 2002

137

100 80 60 40 20 0 0

1800

3600

5400

7200

time (s) thioflavine T fluorescence (% of maximum value)

b 100 80 60 40 20 0 0

3600

7200

10800

14400

2700

3600

time (s)

c thioflavine T fluorescence (% of maximum value)


a

thioflavine T fluorescence (% of maximum value)

articles

100 80 60 40 20 0

0

900

1800

time (s)

under fully aqueous conditions20. These observations suggest that the conditions used in the present study are appropriate for investigating aggregation under conditions where it can be compared directly to the folding process. Aggregation of the wild type and variant AcP molecules was initiated by incubation of each protein at a concentration of 0.4 mg ml–1 in 25% (v/v) TFE, 50 mM acetate buffer, pH 5.5, at 25 °C. By the time of the first acquisition of experimental data, all the samples are completely denatured, as revealed by circular dichroism (CD) spectra and stopped-flow measurements. Within 2 h, wild type AcP converts into aggregates observed by electron microscopy as granules or very short protofibrils ∼4 nm in width11. The aggregates possess extensive β-sheet structure, as indicated by CD and Fourier transform infrared spectroscopy (FT-IR), and have the ability to bind specific dyes such as Congo red and thioflavine T (ThT)11. Formation of these aggregates precedes and correlates well with the development of welldefined amyloid fibrils at longer times for AcP11 as well as other protein systems21,22. Therefore, study of the rate of formation of the first granules and protofibrils enables a relatively rapid comparative analysis of a large number of samples. Only some mutations alter the aggregation rate of AcP The time courses of the increase in ThT fluorescence, the increase in Congo red absorbance at 540 nm and the decrease in the ellipticity at 216 nm were all recorded for several hours after initiating aggregation of the wild type protein (data not shown). In each case, the data were found to be well described by a single 138

Fig. 1 Rate of aggregation for wild type AcP and some representative mutational variants. The proteins are a, wild type AcP (crosses) and I75V (filled squares). b, Wild type AcP (crosses), V20A (empty circles) and Y98Q (filled circles). c, Wild type AcP (crosses), E29D (empty triangles) and A30G (filled triangles). The continuous lines through the data are the best fits to single exponential functions. In preliminary experiments, the aggregation rate of wild type AcP was measured by following the increase of Congo red absorbance, the increase of ThT fluorescence and the decrease of far-UV CD at 216 nm. Analysis of the kinetic traces in the three types of experiment resulted in similar rate constant values of 8.7 (± 1.6) × 10–5, 9.3 (± 0.5) × 10–5 and 9.8 (± 1.6) × 10–5 s–1, respectively. ThT fluorescence was used for studying the aggregation rates of all the AcP variants. The time course of ThT fluorescence was measured seven times for wild type AcP, yielding values of 9.3 × 10–5 for the mean rate constant and 1.6 × 10–5 s–1 for its standard deviation. The latter value was used to assess the significance of the differences between the aggregation rates determined for the various mutants relative to the wild type protein.

exponential function. The kinetic traces were highly reproducible, and the rates determined using any of the three techniques were closely similar (legend to Fig. 1). ThT fluorescence was used for studying the aggregation rates of wild type AcP and the various mutational variants (Fig. 1). All 22 variants with amino acid replacements within the regions of the sequence 1–15 and 32–86 aggregate at a rate which is, within experimental error, that of the wild type protein (ln (vmut / vwt) ∼0) (Fig. 2a). In contrast, all six mutations in the region 16–31 (V17A, V20A, F22L, Y25A, E29D and A30G) and five of the six mutations in the C-terminal region 87–98 (S87T, L89A, Y91Q, S92T and Y98Q) lead to aggregation rates significantly different from that of the wild type protein — that is, they have ln (vmut / vwt) significantly greater or lower than zero. The sensitivity of the aggregation rate to amino acid replacements within these regions indicates that these stretches of the polypeptide chain play critical roles in the rate-determining step of AcP aggregation. In contrast, the lack of effect following substitutions in any other regions suggests that the remainder of the chain is not involved in such a step. From the CD and FT-IR measurements it is clear that a significant proportion of the sequence of AcP is in β-sheet conformation in the amyloid fibrils and even the initially formed aggregates11. Therefore, regions other than those important in the kinetic process examined here are probably part of the β-sheet structure of the aggregates. Moreover, some of the mutations might alter the mechanism of the aggregation process significantly23. Mutations within the P22 tailspike protein, for example, have been found to prevent aggregation by inhibiting entry of a long-lived intermediate into the inclusion body pathway24. The kinetic data presented here, therefore, simply define the regions involved in the rate-determining step of the aggregation process rather than providing details of the mechanism of aggregation or the structure of the resulting aggregates. Table 1 Sequence and solubility of peptides obtained from the sequence of AcP1 Peptide 1–17 18–33 34–53 54–68 69–85 86–98

Sequence STAQSLKSVDYEVFGRV QGVSFRMYTEDEARKI GVVGWVKNTSKGTVTGQVQG PEDKVNSMKSWLSKV GSPSSRIDRTNFSNEKT ISKLEYSNFSVRY

Solubility (mM)2 >10 5.5 ± 0.5 0.014 ± 0.002 >10 >10 0.039 ± 0.004

The six peptides listed in the table encompass the whole sequence of AcP. Peptides are named on the basis of the region of AcP sequence which they span. 2Determined in 50 mM acetate buffer, pH 5.5, room temperature as described (see Methods). When measured in the presence of 25% TFE, the same scale of solubilities was observed. 1

nature structural biology • volume 9 number 2 • febuary 2002

a

3.0

aggregation

2.0

0.0 -1.0

mut w t

/v )

1.0

ln(v

Fig. 2 Change in the rate of a, aggregation and b, folding of AcP resulting from mutation at various positions. Each bar refers to a variant with an amino acid replacement at the position shown on the x-axis. Rate data, reported on the y-axis, are shown expressed as the natural logarithm of the ratio between the rate for the mutant and that for the wild type protein. A value near zero or significantly different from zero implies that the mutant aggregates at a rate similar to or significantly different from that of the wild type protein, respectively. The mutants reported in panel (a) are S5T, V9A, Y11F, V13A, V17A, V20A, F22L, Y25A, E29D, A30G, I33V, G34A, V36A, V39A, T42A, G45A, V47A, V51A, P54A, M61A, W64F, L65V, K67A, P71A, I75V, T78S, E83D, I86V, S87T, L89A, Y91Q, S92T, F94L and Y98Q. Experimental errors of ln (vmut / vwt) were calculated from those of vmut and vwt such that there is a 95% level of confidence that the real values fall within the reported experimental errors51. Hence, values that differ from zero by more than the experimental error represent significant changes of the rate for the variant with respect to the wild type. The data and experimental errors reported in panel (b) are those described32. The ln (vmut / vwt) value of Y94L has been estimated from the free energy change of unfolding and the unfolding rate constant reported32. The ln (vmut / vwt) values of Y91Q, S92T and Y98Q were measured in the present analysis.

-2.0 -3.0 -4.0

10

20

30

40

50

60

70

80

90

70

80

90

residue number

b

2.0

folding 1.0 0.0

mut

wt

/v )

Peptide fragments of AcP have different solubilities -1.0 Different regions of the sequence of AcP have different degrees of β-sheet propensity and -2.0 hydrophobicity (Fig. 3a,b). The two regions found here to be highly sensitive to mutation -3.0 have values of β-sheet propensity and hydrophobicity above the average values calculated for the -4.0 entire sequence. The sequence stretch following the 16–31 region, however, is also highly -5.0 hydrophobic and has a high predicted propensity to form β-sheet structure (Fig. 3a,b). In order to explore this further, the sequence of the protein was dissected into smaller units by synthesizing six peptides that encompass different regions of the sequence and, when taken together, span the entire AcP molecule from the N- to the C-terminus (Table 1). The three peptides corresponding to residues 1–17, 54–68 and 69–85 of AcP were found to be soluble, with no aggregation being detected at concentrations as high as 10 mM in water (Table 1; Fig. 3c). More interesting, these regions have a low hydrophobicity and propensity to form β-sheet structures (Fig. 3a,b). The peptides corresponding to the 18–33 and, in particular, the 86–98 regions of AcP — approximately the 16–31 and 87–98 regions found from the mutational study to be important in aggregation — were found to have significantly lower solubilities (Table 1; Fig. 3c). The peptide corresponding to the 34–53 region of AcP is also very insoluble (Table 1; Fig. 3c). Therefore, a reasonable correlation exists between solubility and the tendency to form hydrophobic interactions and β-sheet structure. In fact, the three regions corresponding to the insoluble peptides all possess a relatively high hydrophobicity and a high tendency to form β-sheet structure (Fig. 3a,c). The six peptides show similar relative solubilities in the presence of 25% (v/v) TFE, although all solubility values are significantly higher in this case. Despite the insolubility of the peptide corresponding to residues 34–53, mutations in this region of the intact AcP molecule did not affect the aggregation rate of the protein. To explore this apparent anomaly between the two types of experiments, CD spectra were acquired for all six peptides, with the exception ln(v


articles

nature structural biology • volume 9 number 2 • february 2002

10

20

30

40

50

60

residue number

of the highly aggregating peptide 87–98 for which the signal-tonoise ratio was not sufficiently high for CD analysis. The spectra were recorded in all cases immediately after extensive centrifugation at 10,000g in order to remove any substantially aggregated material from the samples. The 1–17, 18–33, 54–68 and 69–85 peptides yielded CD spectra typical of a random coil, indicating that these peptides are largely unstructured in aqueous solution (Fig. 4). By contrast, the soluble fraction of the 34–53 peptide has a CD spectrum indicative of a considerable amount of β-sheet structure both in aqueous solution and in the presence of 25% TFE (Fig. 4). Gel filtration chromatography indicated that >90% of the peptide molecules in such a soluble fraction are monomeric. This rules out the possibility that the β-sheet-rich CD spectrum of a saturated solution of this peptide arises from oligomeric forms of the peptide remaining in solution after centrifugation, indicating instead that the monomeric form of this peptide has a large propensity to form β-sheet structure. Comparisons with other systems indicate that this structure probably corresponds to a β-hairpin25. Importance of hydrophobicity and -sheet propensity The sensitivity of the aggregation rate to the mutations within the two regions of the sequence 16–31 and 87–98 provides an opportunity to explore possible reasons for the observed rate changes in terms of the nature of the amino acid substitutions. To assist in this analysis, four further single point mutations at 139

articles

7

6.5

6

5.5

5

0

20

40

60

80

100

residue number

b

40 20 0 -20 -40 -60 -80

0

20

40

60

80

100

80

100

residue number

c 0

20

40

60

residue number

position 20 and two others at position 89 were generated, and the aggregation rates of these variants were measured. We examined, in particular, the changes in the hydrophobicity26, α-helical propensity27 and β-sheet propensity28 predicted to occur as a result of each mutation; all these parameters might be expected to influence the probability of forming aggregates rich in β-structure.The experimentally measured change in aggregation rate was plotted versus the total free energy change predicted as a result of each mutation (Fig. 5a). The latter was calculated as the sum of the hydrophobic effect and the change in the local propensity of the polypeptide chain to convert from an α-helical to a β-sheet conformation (legend to Fig. 5). Despite the simplicity of this approach, a highly significant correlation was found (r = 0.755 and p = 0.0002). Less impressive but significant correlations are also observed when changes of hydrophobicity and secondary structure propensity are considered separately (r = 0.482 and p = 0.0367 for hydrophobicity, and r = 0.708 and p = 0.0008 for secondary structure propensity). The observation that the correlation is more significant when considering the secondary structure term, as opposed to the hydrophobic one, suggests that the former is a more important factor in the analysis. Fig. 4 Far-UV CD spectra of peptides corresponding to various sequence regions of AcP. The peptides are designated by numbers representing the particular regions of the sequence of AcP that they encompass. The CD spectrum of the 87–98 peptide was not recorded due to the low signal-to-noise ratio because of its high insolubility. The CD spectra were acquired in buffer solution. When TFE was added to a final concentration of 25%, the 34–53 spectrum did not show any significant change, whereas the spectra of the other peptides increased in helicity.

140

Fig. 3 β-sheet propensity and hydropathy profiles of AcP. The profiles were constructed using a, the Street and Mayo scale28 and b, the Roseman scale26. Sliding windows of nine residues are used in both cases, except for the N- and C-termini in which narrower windows are used. In panel (a), high numbers indicate large β-sheet propensities. In panel (b), positive and negative numbers indicate hydrophobicity and hydrophilicity, respectively. Values of zero indicate a hydrophobicity equal to a stretch of nine Gly residues. The horizontal lines in the two panels indicate the average β-sheet propensity (a) and hydrophobicity (b) of the AcP sequence. The pale green bands indicate the regions of sequence found to be kinetically important in the conversion of a denatured ensemble of conformations into aggregates (Fig. 2). c, Diagram indicating six peptides spanning various regions of AcP and color-coded to indicate their solubility: brown is low solubility; pale brown, medium solubility; and white, high solubility (Table 1). This pattern of solubilities is observed in both the absence and presence of 25% TFE.

Correlations between ln (vmut / vwt) and changes in α-helical propensity exist when the mutations within the 16–31 and 87–98 regions are considered separately, indicating that this parameter is a relevant factor in both regions. Moreover, a qualitative correlation between α-helical propensity and aggregation behavior has been noted for other protein systems29,30. More important, however, the experimental data for the regions 1–15 and 32–86 of AcP show that the lack of perturbation in the rate of aggregation, which results from mutations in these regions, does not result from a lack of significant changes in hydrophobicity or secondary structure-forming propensities (Fig. 5b). In addition, the observed differences in aggregation rate cannot be attributed simply to the types of mutation. Several of the amino acid substitutions found to have large effects on the aggregation rate in the 16–31 and 87–98 regions of the protein — for instance, Val to Ala, Ser to Thr and Ala to Gly — have not been found to cause significant changes in the remainder of the sequence (legend to Fig. 2). In agreement with our observation that changes of β-sheet propensity upon mutation contribute to changes in the aggregation rate, the insertion of β-sheet-breaker Pro residues has been shown to inhibit aggregation and amyloid formation of the Aβ peptide31. Despite a remarkable correlation (Fig. 5a), some scatter of the data points is evident. Contributions to this scatter will include (i) difficulties in accurately predicting the change of hydrophobicities and secondary structural preferences upon mutation, (ii) specific roles that the side chains might play in the transition from the denatured to the aggregated states, or (iii) the relative importance that the various positions within the 16–31 and 87–98 regions have in the aggregation process. Aggregation and folding involve distinct regions The analysis performed with the peptides obtained from AcP indicates that the regions with an intrinsically high propensity to aggregate are those with a high content of residues that are both hydrophobic and favor the formation of β-structure, notably 10000 18-33 peptide

[Θ] (deg cm2dmol -1)

β -sheet propensity

7.5

hydropathy


a

5000

69-85 peptide

1-17 peptide

54-68 peptide

0 -5000 -10000

34-53 peptide

-15000 -20000

200

210

220

230

240

250

wavelength (nm) nature structural biology • volume 9 number 2 • febuary 2002

regions 18–33, 34–53 and 86–98 (Fig. 3). The mutational study of the intact protein agrees with these findings because the two regions of the sequence important in aggregation (16–31 and 87–98) correspond to two of these peptides, (Figs 2, 3) and because the measured changes of aggregation rate upon mutation correlate with changes in the hydrophobicity and β-sheet propensity of these two regions (Fig. 5). The question naturally arises as to why the region 34–53, which has an intrinsically high propensity to aggregate, does not participate in the rate-determining steps of aggregation of the entire protein (Fig. 2). Clues to a possible answer to this question come from a comparison of the effects of AcP mutations on aggregation and folding, and from the results of a variety of NMR spectroscopic studies of denatured protein ensembles under folding conditions. The mechanism of folding of AcP has been investigated by studying many of the same mutational variants that have been used in the present work to characterize the aggregation process32,33. The two regions of the sequence found from the mutational study to be important in determining the kinetics of aggregation (16–31 and 87–98) approximately correspond, in the native conformation, to α-helix 1 (residues 22–32) and the loop preceding this helix (residues 15–21), and to the short β-strand 5 (residues 94–98) and the long loop preceding it at the C-terminus (residues 86–93) (Fig. 6a). From the analysis previously presented, the two loops 15–21 and 86–93 appear to be highly unstructured in the transition state for folding of AcP — that is, they have low Φ-values32. The short β-strand 5 is also unstructured in the transition state32. The interactions formed in the native state by the side chains of α-helix 1 (residues 22–32) with other regions of the sequence are formed only weakly at this stage of the folding reaction32. Thus, native-like structure in all these regions develops at a late stage in folding32. Within the regions found to be involved in aggregation, the only mutation that causes a large change in the folding rate of AcP is the F94L substitution32,33. This mutation is also the only one within these regions that does not significantly alter the rate of AcP aggregation (Figs 2, 6). In contrast, highly structured regions in the transition state — that is, regions with large Φ-values — cluster around β-strand 1 (residues 7–15), β-strand 3 (residues 47–53) and the preceding turn (residues 42–46) (Fig. 6). Mutations within these regions of the protein have large effects on the rate of folding but little consequence on that of aggregation (Fig. 2). nature structural biology • volume 9 number 2 • february 2002

a

3 2

ln (vmut /vwt )

Fig. 5 Analysis of the changes of aggregation rate of AcP resulting from mutation. Correlation between the change of aggregation rate of AcP upon mutation (ln (vmut / vwt)) and the sum of the calculated changes in the hydrophobicity (∆∆Goctanol-water), propensity of the polypeptide chain to convert from an α-helix to a disordered conformation (∆∆Gcoil-α) and propensity to convert from a disordered conformation to a β-structure (∆∆Gβ-coil) as a result of each mutation. Plots refer to the mutations of residues a, within the two key regions 16–31 and 87–98 and b, within the remaining regions of the sequence 1–15 and 32–86. ∆∆Gcoil-α was calculated using ∆∆Gcoil-α = RT ln (Pαwt / Pαmut), where Pαwt and Pαmut are the α-helical propensities (helix percentages) of the wild type and mutated sequences at the site of mutation, respectively, calculated using the AGADIR algorithm27. ∆∆Gβ-coil was calculated using ∆∆Gβ-coil = 13,640 (Pβwt – Pβmut), where Pβwt and Pβmut are the normalized β-sheet propensities of the wild type and newly introduced residue28, respectively, and 13,640 is the conversion constant from the normalized scale to J mol–1 units. The change of hydrophobicity of AcP at the site of mutation was determined by ∆∆Goctanol-water = ∆Gwt – ∆Gmut, where ∆Gwt and ∆Gmut represent the free energy changes following transfer from water to octanol of the wild type and mutated side chain, respectively (values are from ref. 26). In the first plot (a) the mutations used are V17A, V20A, V20L, V20I, V20S, V20N, F22L, Y25A, E29D, A30G, S87T, L89A, L89T, L89V, Y91Q, S92T, F94L and Y98Q. In the second plot, the mutations used are those listed in the legend of Fig. 2 within the 1–15 and 32–86 regions. A data point for the wild type protein (ln (vmut / vwt) = 0, ∆∆G = 0) is also shown in both plots.

1 0 -1 -2

r = 0.755

-3

p = 0.0002

-4 -20

-15

-10

-5

0

5

10

15

20

∆∆Gcoil− α + ∆∆Gβ− coil + ∆∆Goctanol-water upon mutation (kJ mol-1)

b

3 2

ln (vmut /vwt)


articles

1 0 -1 -2

r = 0.29

-3

p = 0.20

-4 -20

-15

-10

-5

0

5

10

15

20

∆∆Gcoil − α + ∆∆G β− coil + ∆∆Goctanol-water upon mutation (kJ mol-1)

NMR spectroscopy of denatured proteins under folding conditions shows that the denatured ensembles are more compact than predicted by a random coil model because of the inherent attractions between particular types of side chains — for example, those of hydrophobic residues — in different regions of a given sequence34,35. Moreover, the conformations that are populated within such ensembles can be highly native-like36–39. More important, for proteins that fold according to a two-state model, such as the SH3 domain from spectrin and the chymotrypsin inhibitor 2, direct evidence has been obtained that the residual native-like structure in the denatured states under folding conditions closely resembles that of the corresponding transition states for folding40,41. In such denatured species, the regions of the protein involved in the folding nucleus will have less accessibility than expected on a random coil model and, therefore, will have a reduced propensity to form intermolecular interactions. This conclusion provides an explanation for the present finding that AcP residues forming the major interactions within the folding nucleus (such as part of the 34–53 region) are not involved in the rate-determining steps of aggregation. The formation of a β-hairpin by the peptide corresponding to residues 34–53 at concentrations low enough for it to remain soluble and monomeric reinforces this conclusion (Fig. 4). Thus, this region, which is part of the folding nucleus, is probably structured in the denatured ensemble, reducing the probability that it will form the intermolecular interactions that promote the process of aggregation. Underlying principles governing protein aggregation Amino acid substitutions that destabilize the native state of AcP render the protein more prone to aggregation, regardless of the 141

articles


a

b

Fig. 6 Regions of AcP involved in aggregation and folding. The structure of native AcP is shown in both representations with regions in which a mutation changes significantly the rates of a, aggregation and b, folding colored in red. The data for aggregation are taken from those shown in Fig. 2. The data for the folding process are from ref. 32; only regions with high Φ-values (>0.4) are colored in red in panel (b) (see ref. 32 for estimates of average Φ-values of the various regions). The importance of α-helix 2 rises when the main chain hydrogen bonding, rather than the side chain interactions with other residues, is considered47.

position of the mutated residue in the sequence, by increasing the population of unfolded species12. Destabilization of the native state has been recognized as a primary mechanism for natural mutations to exert their pathogenic potential in amyloid diseases17,18. The results described here indicate that certain amino acid substitutions can favor aggregation by acting at a further step in the process — that is, by facilitating the assembly of partially denatured conformations into aggregates. In this regard, it is interesting that all of the six mutations of the prion protein that are responsible for hereditary forms of spongiform encephalopathies, but which leave the conformational stability of the cellular form of the prion protein (PrPC) unaltered42,43, increase either the β-sheet propensity or hydrophobicity of the prion sequence, as predicted by the algorithms used here. The fundamental events in protein folding are dominated by a small number of specific residues involved in the formation of a well-defined, specific folding nucleus that then allows the remainder of the structure to coalesce efficiently around it14,33. By analogy, the present data suggest that, at least for AcP under the conditions studied here, only a small number of residues are involved in the formation of the first intermolecular interactions that nucleate the conversion of the polypeptide chain into an aggregated state. Moreover, the results indicate that there is a kinetic partitioning of folding and aggregation. This finding is particularly remarkable because both processes involve the development of inter-residue interactions of essentially identical character from a common denatured state of the protein. No residues in any of the four long β-strands that form the structural core of AcP have any measurable influence on the rate-determining events of the aggregation process (Fig. 6), although β-strands 2 and 3 are intrinsically very highly aggregationprone. The observation that the propensity of the full length protein to aggregate is reduced by intramolecular interactions associated with the formation of the folding nucleus suggests that the sequences of natural proteins may have evolved such that partially structured states populated during folding have the tendency to resist aggregation under normal folding conditions. This conclusion is also consistent with the proposal that there are specific residues, termed structural gatekeepers, that reduce the aggregation propensity of unfolded proteins44.

needed to assess the generality of this conclusion, this result provides confidence that, similarly to protein folding45, some relatively simple principles also govern the processes of protein aggregation. In general, the formation of stable, ordered aggregates involves, at least in this case, specific residues identified by mutational experiments, offering the prospect of understanding in molecular detail the manner in which the complex process of aggregation is regulated in living systems. Such an understanding should be of particular significance in defining the origins of the debilitating diseases associated with inappropriate deposition of proteins in different types of tissue, leading to the prospect of improved methods of therapeutic intervention. In addition, it should contribute to our understanding of the evolution of the sequences of proteins encoded within the genomes of living organisms and enable us to design more rationally novel proteins that are able to fold efficiently in a given environment.

Conclusions The regions of the AcP sequence responsible for initiating the process of aggregation are those which possess a relatively high hydrophobicity and propensity to form β-sheet structure but which do not participate in the establishment of the folding nucleus. This can be rationalized because such regions are probably solvent-exposed in the denatured ensemble, allowing them to take part readily in the intermolecular interactions involved in aggregation. Although studies on a range of proteins will be

Gel filtration chromatography. Gel filtration chromatography was performed using a Superdex 75 HR 10/30 analytical column (Amersham Pharmacia Biotech) at a flow rate of 0.7 ml min–1. The column was equilibrated with 50 mM acetate buffer, pH 5.5, at room temperature and calibrated by running aprotinin, ribonuclease A, chymotrypsinogen A, ovalbumin and albumin.

142

Methods Production and purification of AcP mutants. Site-directed mutagenesis and purification of AcP mutants were carried out as described46. Synthesis and solubility determination of peptides. Peptides were synthesized and purified as described47. Mass spectrometry confirmed that the purified peptides corresponded to the desired sequences. All the peptides were freeze-dried twice and resuspended in a 50 mM acetate buffer, pH 5.5, at room temperature. The resulting solutions were shaken for prolonged periods of time and subsequently centrifuged at 10,000g for 10 min to remove any large aggregates. The effective solubility of each peptide was estimated by measuring the concentration using appropriate ε280 values calculated for each sequence as described48, except in the case of the 69–85 peptide in which amino acid analysis was used as a consequence of the absence of Tyr, Cys and Trp residues. Circular dichroism. The peptides were dissolved in the presence and absence of 25% (v/v) TFE in 50 mM acetate buffer, pH 5.5, 25 °C at concentrations ranging from 0.01 to 0.2 mM, depending on their solubilities. Far-UV CD spectra of the resulting samples were obtained using cuvettes of 0.1 to 10 cm and a Jasco J-720 spectropolarimeter (UK) thermostated at 25 °C. All the spectra were subtracted by the appropriate background and normalized to mean residue ellipticity.

Aggregation kinetics probed by thioflavine T fluorescence. Aggregation of AcP and its variants was initiated by incubating the protein at a concentration of 0.4 mg ml–1 in 25% (v/v) TFE, and 50 mM

nature structural biology • volume 9 number 2 • febuary 2002


articles acetate buffer, pH 5.5, at 25 °C. Aliquots of 60 µl of this solution, withdrawn at regular time intervals, were mixed with 440 µl of 25 µM ThT and 25 mM phosphate buffer, pH 6.0. The resulting fluorescence was measured using a Shimadzu RF-5000 spectrofluorimeter thermostated at 25 °C. The excitation and emission wavelengths were 440 and 485 nm, respectively. Plots of fluorescence versus time were fit to single exponential functions of the form y = q + A exp (–kx). k values, obtained from the best fits, provided a quantitative measure of the aggregation rate (v) of the various mutational variants. Differences in the aggregation rate between a given variant and wild type AcP are expressed as ln (vmut / vwt), where vmut and vwt are the aggregation rates for the mutated and wild type protein, respectively. Methods based on the measurement of the initial rate of the increase in ThT fluorescence49,50 were found to yield very similar values of ln (vmut / vwt). In all cases, protein concentration was checked immediately before use, using ε280 values calculated as described48. Aggregation kinetics probed by Congo red binding and CD. The formation of aggregates by wild type AcP was also probed by measuring the increase in absorbance of Congo red at 540 nm and the decrease of mean residue ellipticity at 216 nm. For the Congo red experiments, 60 µl aliquots of the protein solutions were with-

1. Fink, A.L. Protein aggregation: folding aggregates, inclusion bodies and amyloid. Folding Des. 3, R9–R23 (1998). 2. Dobson, C.M. Protein misfolding, evolution and disease. Trends Biochem. Sci. 24, 329–332 (1999). 3. Raso, S.W. & King, J. In Mechanisms of protein folding. 2nd edn (ed. Pain, R.H.) 406–428 (Oxford University Press, Oxford; 1999). 4. Leroux, M.R. & Hartl, F.U. In Mechanisms of protein folding. 2nd edn (ed. Pain, R.H.) 364–405 (Oxford University Press, Oxford; 1999). 5. Kelly, J.W. Alternative conformations of amyloidogenic proteins govern their behavior. Curr. Opin. Struct. Biol. 6, 11–17 (1996). 6. Dobson, C.M. The structural basis of protein folding and its links with human disease. Philos. Trans. R. Soc. Lond. B. Biol. Sci. 356, 133–145 (2001). 7. Mattson, M.P., Pedersen, W.A., Duan, W., Culmsee, C. & Camandola S. Cellular and molecular mechanisms underlying perturbed energy metabolism and neuronal degeneration in Alzheimer’s and Parkinson’s diseases. Ann. NY Acad. Sci. 893, 154–175 (1999). 8. Yankner, B.A., Duffy, L.K & Kirschner, D.A. Neurotrophic and neurotoxic effects of amyloid β protein: reversal by tachykinin neuropeptides. Science 250, 279–282 (1990). 9. Varadarajan, S., Yatin, S., Aksenova, M. & Butterfield, D.A. Review: Alzheimer’s amyloid β-peptide-associated free radical oxidative stress and neurotoxicity. J. Struct. Biol. 130, 184–208 (2000). 10. Pastore, A., Saudek, V., Ramponi, G. & Williams, R.J.P. Three-dimensional structure of acylphosphatase. Refinement and structure analysis. J. Mol. Biol. 224, 427–440 (1992). 11. Chiti, F. et al. Designing conditions for in vitro formation of amyloid protofilaments and fibrils. Proc. Natl. Acad. Sci USA 96, 3590–3594 (1999). 12. Chiti, F. et al. Mutational analysis of the propensity for amyloid formation by a globular protein. EMBO J. 19, 1441–1449 (2000). 13. Matouschek, A., Kellis, J.T., Serrano, L. & Fersht, A.R. Mapping the transition state and pathway of protein folding by protein engineering. Nature 340, 122–126 (1989). 14. Fersht, A.R. In Structure and mechanism of protein science. A guide to enzyme catalysis and protein folding. 540–572 (W.H. Freeman & Co., New York; 1999). 15. Grantcharova, V., Alm, E.J., Baker, D. & Horwich, A.L. Mechanisms of protein folding. Curr. Opin. Struct. Biol. 11, 70–82 (2001). 16. Gunasekaran, K., Eyles, S.J., Hagler, A.T. & Gierasch, L.M. Keeping it in the family: folding studies of related proteins. Curr. Opin. Struct. Biol. 11, 83–93 (2001). 17. Hurle, M.R., Helms, L.R., Li, L., Chan, W. & Wetzel, R.A. A role for destabilising amino acid replacements in light chain amyloidosis. Proc. Natl. Acad. Sci. USA 91, 5446–5450 (1994). 18. McCutchen, S.L., Lai, Z.H., Miroy, G.J., Kelly, J.W. & Colon, W. Comparison of lethal and nonlethal transthyrethin variants and their relationship to amyloid disease. Biochemistry 34, 13527–13536 (1995). 19. Ramirez-Alvarado, M., Merkel, J.S. & Regan L. A systematic exploration of the influence of the protein stability on amyloid fibril formation in vitro. Proc. Natl. Acad. Sci. USA 97, 8979–8984 (2000). 20. Main, E.R. & Jackson, S.E. Does trifluoroethanol affect folding pathways and can it be used as a probe of structure in transition states? Nature Struct. Biol. 6, 831–835 (1995). 21. Harper, J.D, Lieber, C.M. & Lansbury, P.T. Jr. Atomic force microscopic imaging of seeded fibril formation and fibril branching by the Alzheimer’s disease amyloid-β protein. Chem. Biol. 4, 951–959 (1997). 22. Conway, K.A., Harper, J.D. & Lansbury, P.T. Jr. Fibrils formed in vitro from αsynuclein and two mutant forms linked to Parkinson’s disease are typical amyloid. Biochemistry 39, 2552–2563 (2000). 23. King, C.Y. Supporting the structural basis of prion strains: induction and identification of [PSI] variants. J. Mol. Biol. 307, 1247–1260 (2001). 24. Betts, S. & King, J. There’s a right way and a wrong way: in vivo and in vitro folding, misfolding and subunit assembly of the P22 tailspike. Structure Fold. Des. 7, R131–R139 (1999). 25. Blanco, F.J, Rivas, G. & Serrano, L. A short linear peptide that folds into a native stable β-hairpin in aqueous solution. Nature Struct. Biol. 1, 584–590 (1994). 26. Roseman, M.A. Hydrophilicity of polar amino acid side-chains is markedly

nature structural biology • volume 9 number 2 • february 2002

drawn at regular time intervals and mixed with 440 µl of a solution containing 10 µM Congo red, 150 mM NaCl and 5 mM phosphate buffer, pH 7.4. The absorbance of such mixtures was measured at 540 nm after 2–3 min of equilibration. For the CD experiments the decrease in ellipticity at 216 nm was followed using a thermostated Jasco J-720 spectropolarimeter (UK) and a cuvette of 1 mm path length. In both cases, fits of the data to single exponential functions proved highly satisfactory.

Acknowledgments We are very grateful for support from the Accademia Nazionale dei Lincei (F.C.), the Fondazione Telethon-Italia (F.C.) and the Wellcome Trust (C.M.D.). The OCMS is supported by the BBSRC, the EPSRC and the MRC. The DSB in Florence is supported by the Italian CNR and the Fondazione Telethon-Italia. The authors thank the Centro Interdip. from the University of Florence for assistance in mass spectrometry; and M. Vendruscolo, K. Plaxco, M. Karplus and A. Fersht for stimulating discussions. Competing interests statement The authors declare that they have no competing financial interests. Received 19 October, 2001; accepted 6 December, 2001.

reduced by flanking peptide bonds. J. Mol. Biol. 200, 513–22 (1988). 27. Lacroix, E., Viguera, A.R. & Serrano, L. Elucidating the folding problem of αhelices: local motifs, long-range electrostatics, ionic-strength dependence and prediction of NMR parameters. J. Mol. Biol. 284, 173–191 (1998). 28. Street, A.G. & Mayo, S.L. Intrinsic β-sheet propensities result from van der Waals interactions between side chains and the local backbone. Proc. Natl. Acad. Sci. USA 96, 9074–9076 (1999). 29. Villegas, V et al. Protein engineering as a strategy to avoid formation of amyloid fibrils. Protein Sci. 9, 1700-1708 (2000). 30. Kallberg, Y., Gustafsson, M., Persson, B., Thyberg, J., Johansson, J. Prediction of amyloid fibril-forming proteins. J. Biol. Chem. 276, 12945–12950 (2001). 31. Wood, S.J., Wetzel, R., Martin, J.D. & Hurle, M.R. Prolines and amyloidogenicity in fragments of the Alzheimer’s peptide β/A4. Biochemistry 34, 724–730 (1995). 32. Chiti, F. et al. Mutational analysis of AcP suggests the importance of topology and contact order in protein folding. Nature Struct. Biol. 6, 1005–1010 (1999). 33. Vendruscolo, M., Paci, E., Dobson, C.M. & Karplus, M. Three key residues form a critical contact network in a transition state for protein folding. Nature 409, 641–645 (2001). 34. Smith, L.J., Fiebig, K.M., Schwalbe, H. & Dobson, C.M. The concept of a random coil. Residual structure in peptides and denatured proteins. Folding Des. 1, R95–R106 (1996). 35. Neira, J.L. & Fersht, A.R. Acquisition of native-like interactions in C-terminal fragments of barnase. J. Mol. Biol. 287, 421–432 (1999). 36. Sinclair, J.F. & Shortle, D. Analysis of long-range interactions in a model denatured state of staphylococcal nuclease based on correlated changes in backbone dynamics. Protein Sci. 8, 991–1000 (1999). 37. Zhang, O. & Forman-Kay, J.D. NMR studies of unfolded states of an SH3 domain in aqueous solution and denaturing conditions. Biochemistry 36, 3959–3970 (1997). 38. Redfield, C., Schulman B.A., Milhollen, M.A., Kim, P.S. & Dobson, C.M. αlactalbumin forms a compact molten globule in the absence of disulfide bonds. Nature Struct. Biol. 6, 948–952 (1999). 39. Dobson, C.M., Sali, A. & Karplus, M. Protein folding: A perspective from theory and experiment. Angew. Chem. Int. Ed. 37, 868–893 (1998). 40. Neira, J.L. et al. Following co-operative formation of secondary and tertiary structure in a single protein module. J. Mol. Biol. 268, 185–197 (1997). 41. Kortemme, T., Kelly; M.J., Kay, L.E., Forman-Kay, J. & Serrano, L. Similarities between the spectrin SH3 domain denatured state and its folding transition state. J. Mol. Biol. 297, 1217–1229 (2000). 42. Swietnicki, W., Petersen, R.B., Gambetti, P. & Surewicz, W.K. Familial mutations and the thermodynamic stability of the recombinant human prion protein. J. Biol. Chem. 273, 31048–31052 (1998). 43. Liemann, S. & Glockshuber R. Influence of amino acid substitutions related to inherited human prion diseases on the thermodynamic stability of the cellular prion protein. Biochemistry 38, 3258–3267 (1999). 44. Otzen, D.E., Kristensen, O. & Oliveberg, M. Designed protein tetramer zipped together with a hydrophobic Alzheimer homology: a structural clue to amyloid assembly. Proc. Natl. Acad. Sci. USA 97, 9907–9912 (2000). 45. Baker, D. A surprising simplicity to protein folding. Nature 405, 39-42 (2000). 46. Taddei, N. et al. Looking for residues involved in muscle acylphosphatase catalytic mechanism and structural stabilization: the role of Asn 41, Thr 42 and Thr 46. Biochemistry 35, 7077–7083 (1996). 47. Taddei, N. et al. Stabilisation of α-helices by site-directed mutagenesis reveals the importance of secondary structure in the transition state for acylphosphatase folding. J. Mol. Biol. 300, 633–647 (2000). 48. Gill, S.C. & von Hippel, P.H. Calculation of protein extinction coefficients from amino acid sequence data. Anal. Biochem. 182, 319–326 (1989). 49. Naiki, H. & Nakakuki, K. First-order kinetic model of Alzhaimer’s β-amyloid fibril extension in vitro. Lab. Invest. 74, 374–383 (1996). 50. Friedhoff, P., von Bergen, M., Mandelkow, E. -M. & Mandelkow, E. A nucleated assembly mechanism of Alzheimer paired helical filaments. Proc. Natl. Acad. Sci. USA 95, 15712–15717 (1998). 51. Taylor, J.R. Introduction to error analysis, the study of uncertainties in physical measurements. (University Science Books, Sausalito, California; 1982).

143