Substrates of Intramembrane Proteases: I Like to Move it, Move it!

Principal Investigator:
Christina Scharnagl

Physics of Synthetic Biological Systems (Technische Universität München) and Chemistry of Biopolymers (Wissenschaftszentrum Weihenstephan für Ernährung, Landnutzung und Umwelt, Technische Universität München)

Local Project ID:
pr48ko, pr92so

HPC Platform used:
SuperMUC of LRZ

Date published:


Intramembrane proteases control the activity of membrane proteins and occur in all organisms. A prime example is g-secretase (GSEC), cleaving the amyloid precursor protein (APP), whose misprocessing is related to onset and progression of Alzheimer's disease (AD). Since a protease's biological function depends on its substrate spectrum, it is essential to study the repertoire of natural substrates as well as determinants and mechanisms of substrate recognition and cleavage. This is the aim of a collaborative research project [1] (FOR2290, DFG Forschergruppe ”Understanding intramembrane proteolysis”, Figure 1). Conformational flexibility of substrate and enzyme plays here an essential role for recognition, complex formation and subsequent relaxation steps leading to cleavage and product release [1,2].

Results and Methods

To uncover functional dynamics of substrates, we employ multi-scale molecular dynamic (MD) approaches that span a range of time and spatial scales. Simulations at atomistic (AT) resolution (CHARMM36 force field, are used to analyze key aspects of structure and conformational dynamics (pr48ko) in the microsecond range. Coarse-grained (CG) models (MARTINI force field, provide the long simulation times and large number of replicas required for a reliable prediction of substrate-enzyme contacts (pr92so). The applied simulation codes, NAMD2.12 ( and GROMACS5 (, are known to be highly scalable on the SuperMUC. Our in silico modeling approach closely connects with the in vitro investigations in order to interpret and guide the experiments, and to validate the simulations.

GSEC cleaves an array of over 90 diverse membrane proteins without showing preferences for specific sequence motifs. As cleavage occurs in the helical transmembrane domain (TMD) of the substrates, the relevance of structural and dynamical features of the substrate's TMD itself for processing seems obvious. We used AT MD simulations to investigate, whether TMDs of substrates join a common intrinsic conformational dynamics differing from the dynamics of non-substrates and disease mutants. Simulations of 2 ms length have been performed for TMD model peptides  (i) in a bilayer of POPC lipids (~42000 atoms), and (ii) in the low-dielectric 2,2,2-trifluoroethanol containing 20 vol-% water (~28000 atoms) mimicking the interior of the enzyme. Using the Sandy-Bridge architecture and 528 cores, we obtained 75 ns/day and 90 ns/day, respectively. The comparison with the results from enhanced sampling (78 runs, aggregate time 15.6 ms) for four TMDs revealed reproducibility of the results from the 2 ms simulations. To our knowledge, the  data collected for 50 TMDs builds the largest database of atomistic MD simulations in the microsecond range. For a subset of model peptides, CG simulations were necessary to determine the impact of the crowed membrane environment used for the solid state NMR experiments (100 peptides, 3000-5000 lipids, 35 water molecules per lipid, 150 ms simulation time, 10 ms/day using Intel-MPI with 700 cores on the Sandy-Bridge nodes). In total, the built-up of the database consumed 45 million core-hours and occupies ~50 TB of disk space.

The challenge was to identify features which provide both, characterization and discrimination of conformational dynamics with high significance. To meet this challenge, we applied a bottom-up approach, investigating (i) local structural and dynamical parameters, (ii) global backbone motions, and (iii) helix  orientation in the membrane. The analyses provides evidence that the sequence diversity of the membrane-embedded part of GSEC substrates translates into a comparable diversity of local and global flexibility, which is also shared by non-substrates. This finding challenges the original assumption [1,2] that substrates are recognized due to an unique pattern of intrinsic backbone flexibility of their TMDs. Ultimately, substrate specificity involves the subtle balance of interactions between substrate, enzyme, and lipids in the crowded cell membrane (Figure 2).

Even though there is no structure of an enzyme-substrate complex available, the redistribution of conformational dynamics in response to  binding-induced stiffening at docking sites can be investigated. The relevant information is obtained by post-processing dynamic cross-correlations of residue fluctuations recorded in the unbound state and allows to scan a large number of experimentally guided interaction models without additional simulations. The scans for the APP TMD [3] revealed that motions targeted by disease mutations are involved in binding-induced relaxations, but differ from the motion favored  in the unbound state (Figure 3). Thus, motions contributing to unbound-to-bound conformational changes might be another key to understand determinants of substrate cleavage.

A major breakthrough was the structure determination of GSEC based on cryo-electron microscopy in 2015. In order to computationally predict substrate-enzyme contact interfaces, we set up the CG approach DAFT (Docking Assay For Transmembrane Components, for the APP TMD-GSEC pair. A typical DAFT run consist of 1000 replicas of a system with ~80000 heavy atoms distributed to 8000 cores and reaches 850ns/day on the Sandy-Bridge nodes. This scaling and the good correlation between in silico predicted and experimentally determined dimerization interfaces of the APP TMD with single GSEC helices provides the basis for the de novo prediction of  binding sites for a variety of substrates.

Recently, domain swap experiments provided evidence for a regulation of the enzyme-substrate assembly by the juxtamembrane regions. Taken together, the results from the MD simulations as well as in vitro experiments suggested that substrate recognition and regulation of cleavage efficiency might be more complicated and shifted the discussion from intrinsic dynamics of substrate TMDs to interactions of full-length substrates determining assembly of the substrate-enzyme complex, that in turn modulates cleavage. These questions were in the focus of the renewal request of FOR2290 granted a three years extension in June 2018.

On-going Research / Outlook

The large number and length of the simulations required in this project, made the use of SuperMUC indispensable. With a production of ~90ns/day for an atomistic system with 28000 atoms, the performance almost doubled compared to 2016 (SuperMUC, Phase 2). In the follow-up project (pr27wa) we will focus on the juxtamembrane substrate domains, their coupling with the TMD, as well as interactions of the full-length substrates with the enzyme – again in close connection to  investigations within FG2290. A large conformational space of the soluble domains makes replicate simulations and a much longer simulations time necessary. Both will benefit from enhanced performance provided by SuperMUC-NG. A main challenge will be the analysis of the data sets across multiple simulations.

References and Links


[2] Dieter Langosch, Christina Scharnagl, Harald Steiner and Marius Lemberg. Understanding intramembrane proteolysis: from protein dynamics to reactions kinetics. Trends Biochem. Sci. 40 (2015), 318-327,  doi:10.1016/j.tibs.2015.04.001

[3] Christina Scharnagl and Alexander Götz. Dissecting conformational changes in APP's transmembrane domain linked to ε-efficiency in familial Alzheimer's disease. BioRxiv (2018), doi:

Research Team

Alexander Götz, Simon Menig, Christina Scharnagl (PI) – all: Technische Universität München (Germany)

Scientific Contact

Dr. Christina Scharnagl
Technische Universität München
Fakultät für Physik
Physik-Lehre Weihenstephan
Maximus-von-Imhof-Forum 4, Raum P051 
D-85350 Freising (Germany)
e-mail: christina.scharnagl [at]

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

LRZ Project IDs: pr 48ko, pr92so

February 2020

Tags: Life Sciences Biochemistry LRZ