The Interaction of Alzheimer's Amyloid-β Peptide With Neuronal Lipid Bilayers

Principal Investigator:
Birgit Strodel

Forschungszentrum Jülich (Germany)

Local Project ID:

HPC Platform used:
SuperMUC of LRZ

Date published:

Amyloid-β (Aβ) peptide oligomers are the major contributing cause of neuronal death in Alzheimer’s disease. To understand how membrane lipids affect Aβ oligomerization, a system that includes six Aβ peptides and a membrane comprised of 1058 lipids was comprised to study these effects using molecular dynamics (MD) simulations. Hamiltonian replica-exchange molecular dynamics HREMD was employed to enhance the configurational sampling afforded by the MD protocol. The aim of this ongoing work is to see how the membrane lipids affect the conformation and morphology of the Aβ oligomers.


Alzheimer's disease (AD) is a neurodegenerative disorder that primarily affects the elderly and is having a greater impact on societies across the globe as the overall life expectancy of humans continues to increase. The exact cause of the neuronal death in AD has yet to be established, though it is widely accepted that the major contributing cause of neuronal death associated with Alzheimer's disease are toxic amyloid-β (Aβ) peptides. Monomeric Aβ (Figure 1) can aggregate into insoluble, relatively inert, rigid structures called fibrils, but also much more toxic, soluble structures of intermediate size, and varying shapes, which are called oligomers. Aβ42 oligomers have been shown to be the most toxic form of the Aβ peptide, though it is still unknown how they originate.

It is becoming increasingly evident that the plasma membrane of neurons plays a role in modulating Aβ aggregation. The aggregation of Aβ was shown to preferentially occur in rigid liquid-ordered phases of lipid membranes that are comprised of sphingomyelin (SM) as well as cholesterol (CHOL). The monosialotetrahexosylganglioside (GM1) has also been found to be involved in Aβ aggregation, however it has been shown both to accelerate aggregation as well as inhibit it. To understand how membrane lipids affect the oligomerization of Aβ, we have comprised a system that includes six Aβ peptides and a bilayer membrane comprised of 1058 lipids (Table 1) and study the aggregation of Aβ42 using molecular dynamics (MD) simulations.

The lipids are distributed symmetrically between the two leaflets of the membrane and provide a good account of the lipids present in neuronal membranes, as they include cholesterol, sphingomyelin and GM1 in physiologically relevant quantities. Moreover, the bilayer is large enough (165 Å x 165 Å surface area) and contains the lipids in sufficient numbers to enable their interactions with Aβ42 to be studied with statistical significance. Moreover, our system contains an experimentally relevant peptide concentration of 2.1 mM in explicit water and in a physiologically relevant NaCl concentration of 150 mM. Therefore, the components of the system under investigation present ideal, physiological conditions.

Results and Methods

We employed Hamiltonian replica-exchange molecular dynamics [1] (HREMD) simulations as implemented in GROMACS [2] and PLUMED [3], along with the Intel® message passing interface (MPI) to carry-out the MD simulations. The all-atom OPLS-AA force field for both the peptides and the lipids was used to model the oligomerization of Aβ42 and its interaction with the aforementioned lipid bilayer.

Ten million CPU-hours were assigned to this project. The HREMD protocol was used to enhance the sampling of the system by assigning a subset of the molecules (the peptides) into a ‘hot region’, and the remainder (membrane and solvent) into a ‘cold region’, where the interactions within the hot and between the hot and cold regions are scaled by a factor λ.

30 replicas of the system were run at increasing temperature, where the coordinates of neighboring replicas could exchange in order to enhance the sampling of the system. 13,440 cores were used for the HREMD job, which generated 150 output files and occupied 1276.4 GB of disk space. A snapshot of the system is included in Figure 2.

In order for us to determine the effect of the bilayer of each oligomer state (monomer through hexamer) we subsequently ran simulations of 1000 ns on each of the six oligomers with the same bilayer in duplicate. The third repeat of each system was completed using computational resources of our project partners in Helsinki. Our subsequent MD runs required 1,792 cores per job, generated 72 output files, and occupied 206.0 GB of disk space.

On-going Research / Outlook

SuperMUC enabled us to employ sufficient computational resources to facilitate the high degree of parallelization necessary for the HREMD protocol to be applied to a system of this size and complexity. Parallelization was also required for the computation of each of the oligomers, and the sheer size of SuperMUC enabled the large amount of computational output to be achieved in a relatively short amount of time.

Our project is still ongoing and especially requires in-depth analysis. After completing the analysis, we will be able to determine which of the lipids are more likely to interact with Aβ, which Aβ conformations are induced by the membrane, which lipids stabilize β-sheets, the peptide conformation most strongly associated with Aβ toxicity.

This is the first Aβ-membrane study on a system of this size and in the presence of a lipid membrane of this many components. Thus, our results should provide new insight into the effect of the neuronal membrane of Aβ oligomerization and membrane-mediated toxicity.

SuperMUC Phase 2 has been an integral part of the progress of our project to date, and we look forward to continuing in this direction in the future. One of the drawbacks of our work is the limit of time scale. In the future we would like to extend the time scale of our work, and we anticipate that SuperMUC-NG will help to make this endeavor possible.

Research Team

Michael C. Owen, Brigit Strodel (PI)

Project Partners

Waldemar Kulig, Ilpo Vattulainen (University of Helsinki, Helsinki Finland)

References and Links

[1] Bussi, G. et al. Mol. Phys. 2014, 112, 379-384.

[2] Pronk, S. et al. Bioinformatics, 2013, 29, 845-854.

[3] G.A. Tribello et. al, Comp. Phys. Comm. 2014, 185, 604

Scientific Contact:

Prof. Dr. Birgit Strodel
Computational Biochemistry Group
Institute of Complex Systems: Structural Biochemistry
Forschungszentrum Jülich
D-52425 Jülich (Germany)
e-mail: b.strodel [@]

NOTE: This report was first published in the book "High Performance Computing in Science and Engineering – Garching/Munich 2018".

LRZ project ID: pr74da

April 2019