Amyloid of the prion domain of Sup35p has an in-register parallel β-sheet structure

EM Indicates Typical Amyloid Morphologies.

Fig. 1 shows EM images of Leu-labeled Sup35NM. The fibrils have the straight, unbranched morphology that is typical of amyloid fibrils. Fibril widths are 9 ± 1 nm, consistent with the fibril diameters on the order of 10 nm previously reported for Sup35NM (7).

Solid-State ¹³C NMR Spectra Indicate a Predominant β-Sheet Structure.

Fig. 2 shows one-dimensional ¹³C NMR spectra of all Sup35NM samples, obtained with magic-angle spinning (MAS). All spectra show a strong line arising from the labeled sites and weaker signals arising from natural-abundance ¹³C nuclei (1.1%). In each spectrum, the line from the labeled sites has an asymmetric shape, most pronounced in the spectra of the Leu-labeled ether precipitate (Fig. 2f) and Ala-labeled fibrils (Fig. 2g) and least pronounced in the spectrum of Tyr-labeled fibrils (Fig. 2a). All line shapes are fit accurately by a sum of two symmetric Gaussian components with unequal areas, chemical shifts, and (in some cases) widths (Table 1).

For the carbonyl NMR lines in spectra of the Tyr-, Phe-, and Leu-labeled fibril samples (Fig. 2 a, c, and d), the major components are shifted upfield from the corresponding random coil chemical shifts, as expected for backbone carbonyl sites in β-strands (32, 40, 41) (Table 1). The minor components are shifted downfield, indicating non-β-strand conformations for a fraction of the labeled residues. The relative areas of the major and minor components suggest that 18 of 20 Tyr residues, six of eight Leu residues, and three of four Phe residues are in β-strands in Sup35NM fibrils.

Hydration of Leu-labeled fibrils (Fig. 2e) results in a reduction of the carbonyl NMR line width from 3.6 ppm to 2.4 ppm, attributable to increased molecular motion in the hydrated state. The major and minor components of the carbonyl NMR line in the spectrum of a Leu-labeled, ether-precipitated sample (Fig. 2f) are reversed relative to those in the spectrum of the Leu-labeled fibrils. The relative areas suggests that three additional Leu residues convert to non-β-strand conformations in the ether-precipitated sample.

For the methyl NMR line in the spectrum of the Ala-labeled fibril sample (Fig. 2g), the major component has a chemical shift approximately equal to the random coil value. The minor component is shifted downfield, as expected for Ala methyl lines from residues in β-strands (32, 40, 41). The relative areas suggest that five of 15 Ala residues are in β-strands, whereas the remaining Ala residues may have non-β-strand conformations.

fpRFDR-CT Data Indicate 0.5-nm Nearest-Neighbor ¹³C–¹³C Distances for Most Labeled Sites.

The fpRFDR-CT technique (39) is an example of a dipolar recoupling technique in solid-state NMR (29, 42), in which radio-frequency pulses are applied in synchrony with MAS to selectively restore (i.e., recouple) the ¹³C–¹³C dipole–dipole interactions that would otherwise be averaged to zero by MAS. The fpRFDR-CT pulse sequence also averages out ¹³C chemical shift interactions. ¹³C NMR signal decays under fpRFDR-CT are then in principle determined only by the recoupled ¹³C–¹³C interactions, which are proportional to the inverse cube of ¹³C–¹³C distances. The fpRFDR-CT data are recorded in a constant-time manner, meaning that the effective dipolar dephasing times in Fig. 3 are incremented by rearrangements of pulses within the pulse sequence while keeping the total pulse sequence length constant. As previously demonstrated (31), signal decays under fpRFDR-CT are therefore insensitive to effects of spin relaxation and pulse imperfections, allowing the determination of relatively long (i.e., >0.5 nm) interatomic distances with ≈10% precision.

Fig. 3a shows fpRFDR-CT data for Leu-, Tyr-, Phe-, and Ala-labeled Sup35NM fibril samples, along with simulated curves for linear chains of ¹³C labels with interatomic distances between 0.4 nm and 0.7 nm. Data for all samples exhibit a component that decays on a time scale of 35 ms, indicating ¹³C–¹³C distances of ≈0.5 nm. Fig. 3b shows fpRFDR-CT data for carbonyl sites in unlabeled Sup35NM fibrils and for the Leu-labeled, ether-precipitated sample. The fraction of rapidly decaying Leu carbonyl NMR signal is reduced significantly in the ether-precipitated sample. The unlabeled fibril sample shows no rapidly decaying fpRFDR-CT signal, as expected if decay on the 35-ms time scale is due to ¹³C–¹³C dipole–dipole couplings.

We estimate the number N_0.5 of ¹³C-labeled sites in each sample that are within 0.5 nm of at least one other ¹³C-labeled site from the relation N_0.5 = N_res(S_iso − S′)/(S_iso − S_0.5), where S′ is the observed fpRFDR-CT signal at 50 ms (approximately equal to 13, 22, 42, and 63 for the Tyr-, Leu-, Phe-, and Ala-labeled samples), N_res is the number of labeled residues (equal to 20, 8, 4, and 15 for the Tyr-, Leu-, Phe-, and Ala-labeled samples), and S_iso and S_0.5 are the expected fpRFDR-CT signals at 50 ms for isolated ¹³C labels and for ¹³C labels that are within 0.5 nm of other labels, respectively. We take S_0.5 = 16 ± 5 and S_iso = 80 ± 10, based on the simulated fpRFDR-CT curves in Fig. 3a and the curve for unlabeled Sup35NM fibrils in Fig. 3b. We obtain N_0.5 = 21 ± 2, 7 ± 1, 2.5 ± 0.5, and 4 ± 2 for the Tyr-, Leu-, Phe-, and Ala-labeled Sup35NM fibril samples. For the Leu-labeled, ether-precipitated sample, N_0.5 = 3 ± 1 (Table 1).

Fig. 3c compares fpRFDR-CT data for an isotopically diluted Tyr-labeled sample with those for the fully labeled sample. Dilution of Tyr-labeled Sup35NM with four parts of unlabeled Sup35NM reduces the fraction of rapidly decaying signal, yielding an apparent N_0.5 = 9 ± 2. This result shows that intermolecular (rather than intramolecular) dipole–dipole couplings are the primary source of signal decay in the fpRFDR-CT measurements.

Solid-State NMR Data Indicate In-Register Parallel β-Sheets in Sup35NM Fibrils.

We assign the 0.5-nm ¹³C–¹³C distances revealed by the fpRFDR-CT data to distances between ¹³C labels in neighboring Sup35NM molecules in β-sheets because the labeled sites (with the exception of Tyr carbonyl labels at residues 45 and 46 and Ala methyl labels at residues 136 and 137) are separated by at least one intervening unlabeled residue and because the distance between neighboring peptide chains in a β-sheet is 0.47 ± 0.01 nm (43). We identify the β-sheets as being of the in-register parallel type because an out-of-register parallel β-sheet or an antiparallel β-sheet would not produce 0.5-nm distances for the large number of sites observed experimentally, especially given the irregular distribution of labeled residues along the Sup35NM sequence. The observation of 0.5-nm intermolecular distances for Ala methyl sites is a particularly strong constraint because a one-residue shift from in-register alignment in an ideal parallel β-sheet would increase the intermolecular distances for Ala methyl sites from ≈0.47 nm to ≈0.67 nm.

Values of N_0.5 derived from the fpRFDR-CT data for Sup35NM fibrils are in good agreement with the numbers of labeled residues in β-strand segments estimated from the ¹³C NMR line shapes (see Table 1). This result supports the interpretation of the fpRFDR-CT data as evidence for in-register parallel β-sheets. The reduced value of N_0.5 in the Leu-labeled, ether-precipitated sample is also consistent with the observation of a stronger non-β-strand component in the ¹³C NMR spectrum.

The value of N_0.5 for the Tyr-labeled sample indicates that nearly the entire N domain of Sup35NM participates in the in-register parallel β-sheet structure, as the Tyr residues are distributed throughout the N domain. The value of N_0.5 for the Phe-labeled sample is then consistent with the fact that the N domain contains three Phe residues. The N domain also contains six Ala residues. The value of N_0.5 for the Ala-labeled sample is also consistent with an in-register parallel β-sheet structure for the N domain, but approximately half of the fpRFDR-CT signal from Ala-136 and Ala-137 (in the M domain) is also expected to decay rapidly because of intramolecular dipole–dipole couplings. Thus, our data suggest that at least one Ala site in the N domain is not in an in-register parallel β-sheet. Ala-42, which immediately follows a proline, is a likely candidate.

Given a 4:1 ratio of unlabeled:labeled Sup35NM molecules in the isotopically diluted Tyr-labeled sample, the probability that a given labeled molecule would have at least one labeled nearest-neighbor in a parallel β-sheet is 0.36. Taking into account the fact that the carbonyl ¹³C NMR signals of Tyr-45 and Tyr-46 should decay because of intramolecular couplings, we predict N_0.5 ≈ 8.5 for the isotopically diluted Tyr-labeled sample, in good agreement with the value derived from the experimental data.

The M Domain Contains In-Register Parallel β-Sheets.

The result that N_0.5 = 7 ± 1 for the Leu-labeled fibrils is surprising, given that only Leu-110 is in the N domain. Apparently, part of the M domain also forms in-register parallel β-sheets. Examination of the Sup35NM sequence suggests that Leu-126, Leu-144, Leu-146, Leu-218, and Leu-238 may be in β-sheets, because none of these residues is in a segment with a high net charge. Leu-212 is likely to be outside of the β-sheet structure, because Leu-212 is preceded by a negatively charged segment and immediately followed by a proline. Leu-154 may also be outside the β-sheet structure. Hydration of Leu-labeled fibrils has little effect on the fpRFDR-CT data (Fig. 3b). This result indicates that β-sheet formation in the M domain is not simply a reversible artifact of the lyophilization of Sup35NM fibrils performed for solid-state NMR measurements.

Additional experiments will be required to identify the β-strands in the M domain unambiguously. Nonetheless, our data for the Leu-labeled sample support the possibility that the M domain adopts a structure comprised of alternating β-strands (forming in-register parallel β-sheets) and loops, with the highly charged and proline-containing segments of the M domain being located in the loops. Similar alternation of β-strand segments and loop (or bend) segments in amyloid fibrils has been established for β-amyloid fibrils (23, 32, 44) and HET-s fibrils (33, 45) and has been suggested for amylin fibrils (46), α-synuclein fibrils (36, 47), and Ure2p prion fibrils (20, 48). The β-strand segments in these fibrils are typically 6–10 residues in length, whereas the loop segments may be only three to four residues in length and can in principle be much longer.

Because our data indicate that nearly all Tyr residues are in β-strands, the loop segments in the N domain of Sup35NM fibrils are likely to be relatively short and the N domain may contain between four and eight β-strands. In contrast, the M domain may contain between three and five β-strands, with the charge-rich residues 130–142, 160–220, and 241–253 likely to be in loops. Fig. 4 shows a very tentative schematic model for the Sup35NM fibril structure suggested by our data.

Comparisons with Previous Structural Studies.

Kishimoto et al. (27) have reported x-ray fiber diffraction patterns of Sup35NM (1–253), Sup35NMp (1–189), and Sup35N (1–123) fibrils in which the meridional 4.7-Å and equatorial ≈9-Å reflections of a cross-β structure are present in dried samples but the equatorial reflections are absent in hydrated samples. They interpreted the absence of equatorial reflections as evidence for a β-helical structure (rather than a layered β-sheet structure), implying that most backbone hydrogen bonds between β-strands would be intramolecular. However, the hydrated samples produced lower total scattering intensities, suggesting that the weaker equatorial reflections may have been obscured by background. Our data are inconsistent with a β-helical structure for Sup35NM fibrils.

Krishnan and Lindquist (26) studied cysteine-substituted Sup35NM derivatives chemically modified with pyrene maleimide (C₂₀H₁₂NO₂). When fibrils were formed from Sup35NM monomers labeled with pyrene maleimide, excimer fluorescence (because of interaction of pyrene pairs) was seen for labels on residues 25–38 (called the head) or residues 91–106 (called the tail) but not the intervening region. Krishnan and Lindquist (26) interpreted this result as being inconsistent with an in-register parallel β-sheet structure but consistent with a β-helical model with head-to-head and tail-to-tail interactions connecting monomers. The size of the pyrene maleimide probe (≈1.0 nm × 0.5 nm, compared with the 0.47-nm distance between β-strands and the 0.8- to 1.0-nm distance between β-sheets in a typical layered β-sheet structure) could have affected the outcome of these studies. In addition, the intensity of pyrene excimer fluorescence depends on the precise geometry of pyrene pair interactions and requires mobility of pyrene labels to achieve the optimal geometry (49, 50). The sterically crowded environment of a fibril core may preclude excimer formation. Parham et al. (51) have reported that deletion of residues 94–112 does not prevent transmission of [PSI⁺].

Sup35NM dimers with cross-links in residues 21–106 exhibited retarded amyloid formation (26), possibly because of formation of an off-pathway dimer structure. Solvent accessibility was greater for cysteines in N than for several cysteines in M (26), suggesting a more compact structure for N. The labeled sites in M could be in loops or solvent-exposed β-sheets. Thus, the data of Krishnan and Lindquist (26) are not inconsistent with our data or the parallel in-register β-sheet model.

Measurements of the mass per unit length of Sup35^1–65 fibrils showed approximately one monomer per 0.47 nm in each of three variants (52). This result is inconsistent with a single-helix β-helical model but is consistent with an in-register parallel β-sheet structure with purely intermolecular backbone hydrogen bonding.

Variants of [PSI⁺], differing in prion stability, in the intensity of the nonsense-suppression phenotype, or in the range of Sup35 prion domain sequences that can propagate them, have been described (10, 53). These phenotypic differences reflect structural variations in the Sup35p fibrils that transmit the trait (11, 12), but the precise nature of structural variations is not known. In the case of β-amyloid fibrils, structural differences between fibril forms include differences in the detailed conformation of the loop segment, differences in certain side-chain interactions, and differences in mass per unit length, but not differences in the identity of β-strand segments or deviations from an in-register parallel β-sheet structure (40). Sup35NM fibrils used in our experiments were produced with initial nucleation at 4°C, a condition reported to favor formation of a single variant of [PSI⁺], but we have no evidence that only a single variant was indeed produced. Future experiments may reveal whether [PSI⁺] variants have different loop segments (in either N or M), different contacts between β-sheets, or other specific structural differences. One intriguing possibility raised by our results is that structural variations in the M domain underlie the phenomenon of [PSI⁺] variants.

Protein Expression, ¹³C-Labeling, Protein Purification, and Fibril Formation.

Sup35NM was expressed for isotope labeling from pJC25NMstop (54) in Escherichia coli strain BL21(DE3) as described (55), with modifications suggested by Ulrich Baxa (personal communication). Sup35NM dialyzed into 5 mM KH₂PO₄ (pH 6.8), and 150 mM NaCl was incubated at 20°C at 1 mg/ml without shaking for 1 week for fibril formation. A dilutely labeled fibril sample was prepared by dissolving Tyr-labeled Sup35NM fibrils and unlabeled fibrils in trifluoroacetic acid in a 1:4 ratio, evaporating the trifluoroacetic acid under nitrogen, dissolving the resulting solid in 6 M guanidine hydrochloride, and allowing fibrils to form under dialysis against 5 mM phosphate buffer (pH 6.8). Ether-precipitated Sup35NM was prepared by dissolving Leu-labeled fibrils in trifluoroacetic acid, diluting 10-fold in cold t-butyl methyl ether, allowing the precipitate to form at −20°C for 24 h, and drying under nitrogen. Details are in supporting information (SI) Text.

Solid-State NMR.

Solid-state NMR measurements were performed at 9.39 T (100.4-MHz ¹³C NMR frequency) and 14.1 T (150.7-MHz ¹³C NMR frequency) using InfinityPlus spectrometers (Varian, Palto Alto, CA) and MAS NMR probes (Varian) with 3.2-mm-diameter rotors. All measurements were at room temperature. ¹³C NMR spectra were recorded at an MAS frequency of 20.00 kHz with ¹H–¹³C cross-polarization (56) and two-pulse phase-modulated ¹H decoupling (57). The fpRFDR-CT measurements were carried out at an MAS frequency of 20.00 kHz under pulse sequence conditions identical to those described previously (20, 31, 40), including pulsed spin-lock detection for an improved signal-to-noise ratio (58). Sup35NM fibrils were lyophilized for solid-state NMR measurements. Sample masses were 2–6 mg. Each fpRFDR-CT data point in Fig. 3 is the result of 384-2560 scans with a 4-s recycle delay.

Raw fpRFDR-CT data were corrected for contributions from natural-abundance (1.1%) ¹³C NMR signals as follows: (i) fpRFDR-CT data were recorded for carbonyl ¹³C signals in an unlabeled sample (Fig. 3b) and fit to a linear decay, resulting in a standard natural-abundance signal S_na(t) = 100 − 0.392t, where t is the effective dipolar dephasing time in milliseconds; (ii) the average number N_na of natural-abundance ¹³C nuclei per Sup35NM molecule was evaluated as 0.011(N_sites − N_labels), where the number N_sites of carbon sites potentially contributing to the fpRFDR-CT signals was taken to be 347 for carbonyl signals (for measurements on the Leu-, Tyr-, and Phe-labeled samples) and 68 for methyl signals (for measurements on the Ala-labeled sample) and the number N_labels of ¹³C-labeled sites per molecule was taken to be 8, 20, 4, and 6 for the Leu-, Tyr-, Phe-, and Ala-labeled samples (taking into account the 40% incorporation of ¹³C in the Ala-labeled sample). N_labels was taken to be 4 for the isotopically diluted Tyr-labeled sample; (iii) the corrected fpRFDR-CT data S(t) were calculated from the raw data S_raw(t) according to the equation S(t) = [S_raw(t) − f_naS_na(t)]/(1 − f_na), with f_na = N_na/(N_na + N_labels), after scaling the raw data so that S_raw(0) = 100. Corrected data are plotted in Fig. 3. Data for the Ala-labeled sample in Fig. 3a are additionally corrected for the 40% incorporation of ¹³C by adding 0.36 N_labels = 2.16 to N_na, based on the fact that a fraction (1 − x)² of the ¹³C labels would not have a nearest-neighbor ¹³C label within 0.5 nm in an in-register parallel β-sheet if the fractional incorporation were x.

Simulations of fpRFDR-CT data were performed as described previously (31). These simulations assume a linear chain of equally spaced ¹³C labels, appropriate for an ideal in-register parallel β-sheet with either backbone carbonyl or side-chain β-carbons.