Conformational Conversion during Amyloid Formation at Atomic Resolution

Summary Numerous studies of amyloid assembly have indicated that partially folded protein species are responsible for initiating aggregation. Despite their importance, the structural and dynamic features of amyloidogenic intermediates and the molecular details of how they cause aggregation remain elusive. Here, we use ΔN6, a truncation variant of the naturally amyloidogenic protein β2-microglobulin (β2m), to determine the solution structure of a nonnative amyloidogenic intermediate at high resolution. The structure of ΔN6 reveals a major repacking of the hydrophobic core to accommodate the nonnative peptidyl-prolyl trans-isomer at Pro32. These structural changes, together with a concomitant pH-dependent enhancement in backbone dynamics on a microsecond-millisecond timescale, give rise to a rare conformer with increased amyloidogenic potential. We further reveal that catalytic amounts of ΔN6 are competent to convert nonamyloidogenic human wild-type β2m (Hβ2m) into a rare amyloidogenic conformation and provide structural evidence for the mechanism by which this conformational conversion occurs.

(A) Representation of the hydrophobic (green) and electrostatic (blue/red) surfaces of Hβ 2 m and ΔN6 (lowest energy structures) using APBS (Baker et al., 2001) and Pymol (DeLano, 2002). Highlighted residues are Phe30, Pro32, Phe62, Leu64, Phe70 and His84 (sticks) that show some of the largest movements of sidechain orientation in the two structures (see also Figure 3C). (B) 15 N longitudinal relaxation (R 1 =1/T 1 ) and { 1 H} 15 N nOe relaxation measurements of 500 μM Hβ 2 m (black) and 500 μM ΔN6 (red) at pH 7.5, 25 °C. Circles highlight residues for which no transverse relaxation rate could be determined due to resonance overlap, line broadening or the residue being a proline. Black crosses mark missing assignments. Rainbow coloured ribbons above indicate the secondary structure content of Hβ 2 m and ΔN6 deduced from the final set of 30 lowest energy structures using DSSPcont (Carter et al., 2003). The error was estimated using duplicates.
(C) Cartoon overlay showing one monomer of the hexameric β 2 m taken from the crystal structure of H13F (3CIQ, in blue) (Calabrese et al., 2008) and the lowest energy structure of ΔN6 (red). Pro32 (blue and red sticks, spheres, respectively) and the disulfide bond (Cys25-Cys80, sticks) are highlighted.
(D) RMS Cα [Å] of the overlay of the structures shown in (C). The rainbow coloured ribbon indicates the secondary structure elements of ΔN6 deduced from the final set of 30 lowest energy structures using DSSPcont (Carter et al., 2003).
(E) Overlay of the structures of H13F and ΔN6 showing residues whose sidechains differ most significantly (> 2 Å) in orientation in the two structures shown in (C).  (C) Hydrogen exchange kinetics of 80 μM Hβ 2 m or ΔN6 measured at pH 7.2, 37 °C or pH 7.2, 25 °C, respectively. The intrinsic rates of HX (k c ) were calculated as described (Bai et al., 1993). Measured rates are denoted k ex . The β-strands in native Hβ 2 m and ΔN6 are shown above in rainbow colour. The rate of HX of ΔN6 is too fast to measure by NMR at 37 °C.
The error was estimated from the noise level of the experiment.     The residues listed show the most intense long-range nOes for both constructs that define the arrangement of residues highlighted in Figure 3C. Note that a full annotation of chemical shifts and peak lists is deposited in BMRB with accession numbers 17165 and 17166 for Hβ 2 m and ΔN6, respectively.

Supplemental Experimental Procedures
Assembly of amyloid fibrils. For all experiments a protein aliquot stored in 10 mM sodium phosphate buffer, pH 7.2 at -80 °C was thawed on ice and centrifuged (10 min, 10,000 g, 10 °C The soluble fraction obtained after centrifugation (14,000 g, 10 min) was analysed by SDS-

PAGE.
Negative-stain EM. Carbon coated copper grids were prepared by the application of a thin layer of formvar with an overlay of thin carbon. Samples were centrifuged (14,000 g, 10 min) and the pellets resuspended in deionised water and then applied to the grid in a drop-wise fashion. The grid was then carefully dried with filter paper before it was negatively stained by the addition of 18 µl of 2 % (w/v) uranyl acetate. Micrographs were recorded on a Philips CM10 electron microscope at moderate dose (~ 100 e Å -1 ).
NMR spectroscopy and structure determination. Sequential assignments were obtained from analysis of HNCA, HNCO, HN(CO)CA, CBCA(CO)NH, HNHA and 1 H-15 N NOESY-HSQC (Vuister and Bax, 1993;Kay et al., 1994;Muhandiram and Kay, 1994;Zhang et al., 1994;Zhang et al., 1997). Spectra were processed using NMRPipe (Delaglio et al., 1995) and analysed in CCPN analysis (Vranken et al., 2005). Aliphatic sidechain resonances were assigned on the basis of H(C)CH-TOCSY, (H)CCH-TOCSY and (H)CCH-COSY (Bottomley et al., 1999). Aromatic specific sidechain assignments were made using 1 H-13 C CT HSQC, HB(CBCGCD)HD and HB(CBCGCDCE)HE spectra (Yamazaki et al., 1993) and short Hβ-Hδ nOes. For the measurement of residual dipolar couplings 1 H-15 N J-modulated HSQC spectra (Tjandra et al., 1996) were acquired in the presence and absence of 7 mg ml -1 or 15 mg ml -1 pf1 bacteriophage for Hβ 2 m and ΔN6, respectively. NOe distance restraints were derived from 120 ms three-dimensional 1 H-15 N NOESYHSQC, three-dimensional 1 H-13 C NOESY-HSQC (Muhandiram et al., 1993;Smallcombe et al., 1995) and threedimensional aromatic 13 C filtered NOESY spectra. Torsion angles phi and psi were predicted from 1 Hα, 13 Cα, 13 Cβ, C' and backbone 15 N chemical shifts using TALOS (Cornillescu et al., 1999). NOes were assigned and structures calculated in a two stage process. In the initial stage structures were calculated using the Marvin/PASD simulated annealing protocol (Kuszewski et al., 2004) from X-PLOR-NIH v2.17 (Schwieters et al., 2006) with all measured nOe peaks and TALOS restraints. In the second stage the 50 structures with the lowest energy were then transferred into the first round of an ARIA 2.1 calculation (Nilges et al., 1997) along with high probability assignments. The RDC alignment magnitude and rhombicity were calculated as an average from the initial 50 structures from the first stage with the DC utility from NMRPipe (Delaglio et al., 1995) and RDCs were used as variable angle restraints (VEAN) and SANI restraints  with a force constant of 0.1.
The final structures were refined in a water box using standard ARIA parameters, the length of the two slow cooling stages in the ARIA protocol were extended by a factor of four as described (Fossi et al., 2005). Network anchoring was switched on throughout the calculation (Linge et al., 2004). During refinement the nOe distance restraint network was corrected for spin diffusion. The adjustment is based on the calculation of a theoretical intensity matrix from the set of structures produced each iteration. The theoretical intensity values were then used to calibrate the experimental volumes, and to correct the distance target. The calibrated volumes could then be used to estimate the error. All ARIA calculations were carried out using CNS 1.1 (Brunger et al., 1998). The final structure ensemble of native Hβ 2 m and ΔN6 was based on a total of 2065 or 2565 experimental nOe restraints, 128 or 118 dihedral angle restraints and 75 or 76 1 H-15 N residual coupling restraints, respectively (Table 1, Table S2 and Figure 3). Structures were validated with WHAT-CHECK (Hooft et al., 1996) and PROCHECKNMR (Laskowski et al., 1996). The molecular structure figures were generated using MolMol (Koradi et al., 1996) andPymol (DeLano, 2002).
SOFAST-HMQC NMR experiments were carried out as described (Schanda and Brutscher, 2005). For refolding experiments: Hβ 2 m was denaturated in 8 M urea at 37 °C prior to refolding via 10-fold dilution with 25 mM sodium phosphate buffer pH 7.5 at 25 °C and the sample was then placed immediately into the NMR spectrometer. The first spectrum was recorded approximately 2 min after refolding was initiated. The length of each experiment was between 30 s and 5 min, d1 delays were between 0.500 and 0.550 seconds and the number of scans collected was between 2 and 8. Chemical shift referencing of all samples was carried out using DSS as an internal standard for 1 H. 15 N and 13 C were indirectly referenced. native Hβ 2 m adopted variable conformations in the 30 lowest energy structures calculated for ΔN6 with only 32 % and 0 % of molecules adopting -strand conformation for these residues, respectively. By contrast in native Hβ 2 m residues 50 and 51 establish both 53 % βstrand structure. β-strand C' in native Hβ 2 m has a likelihood of 100 % whereas the likelihood in ΔN6 is reduced to 74 %. The 3 10 -helix for residues 32-34 probably stabilises the transisomer of X-Pro32 and has a likelihood of 100 % in ΔN6 whereas the likelihood in native Hβ 2 m is 0 %.

N NMR relaxation experiments.
Backbone 15 N transverse relaxation (R 2 =1/T 2 ), 15 N longitudinal relaxation (R 1 =1/T 1 ), { 1 H} 15 N nOe relaxation measurements were carried out as described (Farrow et al., 1994). Duplicate measurements and spectral noise levels were used to obtain an estimate of the error. The R 2 relaxation measurements of all constructs were performed at 500 MHz using a series of 10 experiments with relaxation delays ranging from 16.512 ms to 165.12 ms. The R 1 relaxation measurements of all constructs were performed at 500 MHz using a series of 11 experiments with mixing times ranging from 0 s to 1.28 s. For { 1 H} 15 N nOe relaxation time experiments amide protons were pre-saturated with 120 ° pulses for 3.5 s prior to the experiment. All relaxation measurements were performed using 80-500 µM protein in 81-89.5 mM NaCl (giving a total ionic strength of 100 mM), 10 % (v/v) D 2 O and 0.02 % (w/v) sodium azide in 10 mM sodium phosphate buffer, pH 6.2-8.2.
Hydrogen Exchange NMR. All samples were adjusted to pH 7.2 or 6.2 in 10 mM sodium phosphate buffer and freeze-dried prior to dissolving them in 85 or 89.5 mM NaCl (to a total ionic strength of 100 mM), 100 % (v/v) D 2 O and 0.02 % (w/v) sodium azide. The dead-time of the experiment was approximately 5-10 min and data acquisition (5-15 min) was carried out using SOFAST-HMQC NMR methods (Schanda et al., 2005). The data obtained were fitted (Origin, Originlab © ) to a single exponential and the error was estimated from the noise level of the experiment.