Structural insights into peptide self‐assembly using photo‐induced crosslinking experiments and discontinuous molecular dynamics

Abstract Determining the structure of the (oligomeric) intermediates that form during the self‐assembly of amyloidogenic peptides is challenging because of their heterogeneous and dynamic nature. Thus, there is need for methodology to analyze the underlying molecular structure of these transient species. In this work, a combination of fluorescence quenching, photo‐induced crosslinking (PIC) and molecular dynamics simulation was used to study the assembly of a synthetic amyloid‐forming peptide, Aβ16‐22. A PIC amino acid containing a trifluormethyldiazirine (TFMD) group—Fmoc(TFMD)Phe—was incorporated into the sequence (Aβ*16–22). Electrospray ionization ion‐mobility spectrometry mass‐spectrometry (ESI‐IMS‐MS) analysis of the PIC products confirmed that Aβ*16–22 forms assemblies with the monomers arranged as anti‐parallel, in‐register β‐strands at all time points during the aggregation assay. The assembly process was also monitored separately using fluorescence quenching to profile the fibril assembly reaction. The molecular picture resulting from discontinuous molecule dynamics simulations showed that Aβ16‐22 assembles through a single‐step nucleation into a β‐sheet fibril in agreement with these experimental observations. This study provides detailed structural insights into the Aβ16‐22 self‐assembly processes, paving the way to explore the self‐assembly mechanism of larger, more complex peptides, including those whose aggregation is responsible for human disease.

assemblies. [1][2][3][4] Consequently, shorter synthetic peptide fragments from longer amyloidogenic sequences offer convenient model systems with which to explore peptide self-assembly. 5,6 A case in point is the Gly-Asn-Asn-Gln-Gln-Asn-Tyr (GNNQQNY) sequence, taken from the prion-determining domain (PrD) of the 635 residue Sup35 yeast protein. 7, 8 Eisenberg and co-workers established in 2001 that in aqueous conditions this sequence self-assembles into highly ordered fibril structures that display the characteristic cross-β X-ray diffraction pattern of amyloid. 6 The short nature of GNNQQNY enabled the formation of micro-crystals suitable for electron diffraction, which demonstrated that in the crystals the peptide had formed a parallel in-register β-sheet structure involving a steric zipper with a dry interface between the interdigitated side-chains. 7 GNNQQNY clearly demonstrates that peptide fragments can be useful models of longer amyloid sequences, revealing atomic level information about short segments that may be relevant to the assembly of their longer peptide counterparts.

| TEM analyses
Samples were prepared and analyzed by TEM as described previously. 25 Briefly, TEM images were taken at the stated time points and at the end of each experiment by removing 5 μl from the assembly reaction and incubating this sample on carbon-formvar grids for   (1): where A is the determined calibration constant, z is the charge state of the ion, B is the exponential factor (determined experimentally), t D is the corrected absolute drift time, m ion is the mass of the ion and m gas is the mass of the gas used in the ion-mobility cell (N 2 ). Data were processed using MassLynx v4.1 and Driftscope software supplied with a mass spectrometer.

| DMD simulation
In this work, three independent DMD/PRIME20 simulations were carried out for 12 μs in the canonical (NVT) ensemble. The two major nonbonded interactions of the PRIME20 model are the directional square-well backbone hydrogen bonding interaction and the nondirectional square-well potential interaction between two sidechain beads. The potential energy parameters between the 20 different amino acids include 210 independent square-well widths and 19 independent square-well depths derived by using a perceptron learning algorithm that optimizes the energy gap between 711 known native states from the Protein Data Bank and decoy structures. 31 The Andersen thermostat was implemented to maintain the simulation system at a constant temperature. 32 One hundred and ninety-two peptides were initially randomly placed in a cubic box with a length of 321.0 Å, corresponding to a peptide concentration of 10 mM. The reduced temperature is defined to be T * = k B T/ε HB , where ε HB = 12.47 kJ/ mol is the hydrogen bonding energy. The reduced temperature T * of the simulations was set to be 0.193, which corresponds to 326 K in real temperature units. 33 3 | RESULTS AND DISCUSSIONS

| Analysis of cross-linked products
Cross-linking studies focused on homomeric assembly reactions of Aβ* [16][17][18][19][20][21][22] , that is, every peptide in the sample contained a TFMD group at position Phe20. On photolysis, (TFMD)Phe produces a highly reactive carbene which can insert rapidly and indiscriminately into proximal bonds to form a permanent covalent link. 28,35 As we have shown previously, such cross-linking approaches can yield information on noncovalent organization in self-assembled structures. 26 Figure S1C) fibrils are formed. The resulting mass spectrum identified five major species as photolysis products ( Figure S1c).
Importantly, the IMS function on the mass spectrometer can provide additional information as it facilitates separation of ions with similar m/z ratios; cross-linked monomer, dimer and small amounts of trimer could be observed within the sample, in agreement with our previously published studies ( Figure S1B). 26,27 Tandem MS/MS sequencing fragments a peptide along its amide backbone, producing a variety of different fragment ions. 36 Information for further alternatives). Moreover subtle differences in reactivity preference of the carbene arising from diazirine photolysis could also lead to preferential reaction with Lys16 rather than Val18.

| Aβ 16-22 forms anti-parallel, in-register β-sheets in early protein assemblies
The mass spectrum of cross-linked products from Aβ* 16
The simulations were conducted as described in Section 2 and a series of simulation snapshots were taken at different time points, as shown in Figure 6. After 652 ns of simulation time, most peptides were still in a random coil conformation with some disordered aggregates and small amounts of ordered oligomers present. As the simulation prog- The kinetics of Aβ [16][17][18][19][20][21][22] aggregation can be measured in the simulation by calculating the percentage of residues that are in a β-sheet conformation at different simulation times. Figure 7 demonstrates that, in agreement with the fluorescence quenching data, aggregation proceeds via a rapid increase in β-sheet content, as measured by the decrease in the system potential energy (from 0 to 50% in the first 4,000 ns) followed by a slower second phase (4,000-12,000 ns). The TEM time course (Figure 1) and the simulation snapshots (Figure 6; i.e., hydrophobic collapse and the presence of small amounts of fibrils at early time points followed by a transition to/continued fibril formation) are concordant, as is the two phase kinetics observed in both the fluorescence quenching and the simulations.

| Side-chain contacts formed during DMD simulations of Aβ 16-22 self-assembly
The DMD/PRIME20 simulations approach taken in this study can also be used to assess the distance between the beads which represent specific side-chains beads as the simulation progresses and may be able to identify structures that cannot be resolved in the PIC experiment. To do this, the nearest interpeptide sidechain contacts for all Phe19 and Phe20 residues in the system are calculated and shown in Figure 8. Figure S2 shows a schematic illustrating the simplest antiparallel/ parallel in-register/out-of-register potential arrangements that can be envisioned. At the simulation end point, both Phe19 and Phe20 make the most contacts with the residues that are directly opposite their side-chains in an anti-parallel, in-register orientation (Phe19 and Val18, respectively, see Figure 3). Although in isolation the Phe19-Phe19 contact observed in the simulation would be consistent with both parallel and anti-parallel orientations of monomers as in-register β-strands, in combination with the Phe20 data, the identified contacts are most consistent with an anti-parallel, in-register orientation as the dominant mode of interaction. The discrepancy between these MD simulations and the PIC experiments where a cross-link between Lys16 and Phe20 dominates is reconciled by the fact that PRIME 20 models amino acid side chains as spheres, thus such rotational preferences are not resolved and, instead, a simple distance relationship is observed. Within the data, three distinct phases are observed, which occur on the 0-2,000, 2,000-4,000, and 4,000-12,000 ns timescales. As can be seen in Figure 8a, the second most frequent contact that residue Phe19 makes during the simulation is with Val18 (curve labeled Phe19-Val18). This contact increased steadily until 2,000 ns, at which point no further increase occurred.
Contact between Phe19 and Val18 would be possible in a parallel out-of-register conformation ( Figure S2). When analyzing the contacts that Phe20 forms during this initial period (Figure 8b), two significant F I G U R E 4 At all time points analyzed Aβ* [16][17][18][19][20][21][22] forms both intra-and intermolecular peptide cross-links. Products from the reaction between the carbene and H 2 O are labeled in blue and peptide cross-links are labeled in black [Color figure can be viewed at wileyonlinelibrary.com] contacts (other than the dominant Phe20-Val18 contact) are observed: Phe20-Leu17 and Phe20-Phe20. Phe20-Leu17 interaction would not occur for any of the simplest monomer organizations; this contact again continued to increase in population until 2,000 ns at which time the number of contacts remained stable for the rest of the simulation. The Phe20-Phe20 contact, possible in a parallel-in-register alignment ( Figure S2), increased at a slightly slower rate than the Phe20-Leu17 contact in the first phase. However, rather than stopping at 2000 ns, it continued to increase until it plateaued at 4,000 ns. Taken together, the results suggest that as the simulation progresses, an anti-parallel in-register β-sheet forms, while a plethora of alternative structures are also present including disordered aggregates (particularly at early time points), where a heterogeneous distribution of sidechain contacts would be expected. After 2000 ns, however, the anti-parallel, in-register alignment starts to dominate, while the contacts for other alignments plateau, indicating that these structures no longer grow, and may interconvert to the in-register alignment, in agreement with the experimental observations made by Lynn and co-workers. 13 These series of simulations demonstrate the power of combining F I G U R E 5 Annotated tandem MS/MS spectra demonstrating that at all time points analyzed Aβ* [16][17][18][19][20][21][22] forms anti-parallel, in-register β-strands. In the spectra, b ions are highlighted in red, y ions in green and any double fragmentation products in blue [Color figure can be viewed at wileyonlinelibrary.com] simulations and experimental data to gain a fuller picture of peptide self-assembly at an atomistic level.

| Proposed mechanism of aggregation and comparison with previously proposed Aβ 16-22 aggregation mechanisms
According to the experimental data, the following mechanism for Aβ [16][17][18][19][20][21][22] self-assembly is proposed (Figure 9): peptides aggregate rapidly, forming both fibrils and small amounts of amorphous aggregates (the initial decrease in fluorescence, 5 min). These amorphous aggregates then form fibrils, with most of the self-assembly reaction completed within 1-2 hr (as evidenced by the plateau in the fluorescence quenching data). The fibrils formed at these time points tend to be isolated and unbundled. As the self-assembly reaction continues, the fibrils start to bundle together and coalesce, forming dense mats of fibril structures (after 2 weeks). These large mats (partly) exclude H 2 O, forming a series of dry interfaces, in turn reducing the opportunity for H 2 O to quench the carbene and promoting the formation of interpeptide cross-links.
Lynn and co-workers previously observed a transition from outof-register, anti-parallel β-sheets before Aβ [16][17][18][19][20][21][22] assembles into its final in-register alignment. 13 No such transition was discernable in our cross-linking data. There could be a number of reasons for this: 1. The out-of-register alignment may be lowly populated compared with the in-register alignment and as such may not be captured by the cross-linking experiments at early time points; this is supported by our simulations.