Alzheimer's amyloid-β and the disordered structural ensemble characterized using molecular dynamics and NMR spectroscopy
We used simulations and NMR experiments to investigate the diverse structure of amyloid-β (Aβ) peptide in the soluble non-aggregated form in order to better understand this peptide's role in Alzheimer's disease. Because amyloid-β is intrinsically disordered in its monomeric state, the combination of molecular dynamics simulation and NMR spectroscopy was crucial to determining the individual conformations that make up the amyloid-β structural ensemble. Initially we focused on amyloid-β 1-42 (Aβ42), which is the most toxic form of amyloid-β. We collected homonuclear Nuclear Overhauser Effect (NOE) data on the peptide, and used extensive molecular dynamics simulations to characterize its conformational ensemble. We found that the conformational ensemble of Aβ42 is extremely heterogeneous. However, it also contains many structured populations with long-range NOE contacts. This is in contrast to Aβ21-30, an amyloid-β fragment. Aβ21-30 is mostly extended and unstructured, with no long- range NOEs measured. Next we characterized Aβ40, another common form of amyloid-β, which is less toxic and aggregation prone than Aβ42. Again we saw many long-range NOEs and structured conformations in the Aβ40 ensemble, but the most populated conformations for Aβ40 and Aβ42 were quite different. From our simulations we had seen that Aβ42 adopts a β-turn and β-strand, which together form the most common long-range interaction of the peptide, and that this turn is consistent with the same bend and β-strand segment seen in the aggregated form of the peptide. Aβ40 also adopts many different long-range β-strand conformations, however, none of them are similar to the fibril-like turn and β-strand seen in the Aβ42 ensemble. This is one possible explanation for the greater aggregation rate and toxicity of Aβ42.
Amyloid-β presents a difficult case for characterizing an intrinsically disordered disease protein because it contains many structured conformations within its ensemble. We therefore decided to examine the effectiveness of different computational methods for determining the conformational ensemble of this intrinsically disordered protein. We compared the knowledge- based approach to our de novo molecular dynamics approach. The knowledge-based approach randomly generates an ensemble and refines it to fit the NMR data. The de novo molecular dynamics approach, on the other hand, uses no experimental information to form the amyloid-β ensemble. In both methods, we compare the simulated ensemble to the experimental data after it is created. We found that the knowledge-based approach is highly dependent on the starting pool of structures that it refines, and that a randomly generated pool does not contain structured conformations which are able to fit the NMR data. We also found that certain types of NMR data, like J-coupling constants and NOEs, do a much better job of distinguishing between vastly different ensembles than other types of NMR data like chemical shifts, which are calculated to be the same for both unstructured and heterogeneous structured ensembles. We did find that the knowledge-based approach was useful for further refining the molecular dynamics simulation ensemble to give a better fit to the NMR data. This refinement yielded a slightly different picture of the Aβ40 and Aβ42 monomer conformational ensembles. The refined Aβ42 ensemble still contains the fibril-like turn and β-strand as its major feature, but in the refined Aβ40 ensemble we see many fewer β-strands than in the molecular dynamics ensemble. Our revised picture of the two peptides shows that Aβ40 is less structured than Aβ42, with the most populated β-strand of Aβ40 forming near its N-terminus. Aβ42, with two additional residues at the C-terminus, forms more C-terminal hydrophobic interactions, often adopting a large loop that nucleates a fibril-like turn and β-strand near the middle of the peptide sequence. Thus, the Aβ42 C-terminus does not form a β-strand itself, but promotes β-structure at a different region of the sequence, while preventing the type of β-strands formed in the Aβ40 ensemble.
After fully characterizing the amyloid-β monomer ensemble, we were interested in studying an oligomer of amyloid-β, which is believed to be the toxic agent in Alzheimer's disease. In collaboration with the Schaffer group, we assessed the toxicity of an Aβ42 oligomer, known as the globulomer, on cultures of human cortical neurons. This oligomer, which can be prepared consistently and does not aggregate to form fibrils, was found to induce neuronal cell death, indicating that it could be a toxic complex of amyloid-β. This led us to an investigation of the Aβ42 globulomer structure, known to consist of β-sheets. One proposed model of the globulomer is based on NMR data from a small globulomer precursor. Another model of the globulomer derives from coarse grain simulations of amyloid-β prefibrils. We used molecular dynamics simulations to begin a comparison of these two models. Based on our preliminary simulations, the prefibrillar model seems to maintain a more stable β-sheet structure than the NMR-based model. However, so far the NMR-based model has only been simulated as a dimer unit, and may be more stable when more chains are added. We have also calculated NMR observables from each of the two models and we find that J-coupling and amide exchange experiments may be useful in determining which model more accurately represents the globulomer. Future NMR experiments as well as calculation of NOE data from the simulations will help to form a better picture of this toxic Alzheimer's oligomer. Like the amyloid-β monomer, the oligomer may occupy a range of conformational states that form a diverse ensemble, and therefore molecular dynamics simulations as well as NMR data are crucial to fully representing its structure.