LIFE SCIENCES

Results and Methods

We applied MD simulations to study structural transitions in the insulin receptor ectodomain (IR-ECD) – initially based on crystallographic data that were available at that time (protein database structure entry (PDB) ID: 4ZXB) [2]. The simulated IR ectodomain system consisted of about 1 million atoms in a simulation box of 22.4 x 22.4 x 19.9 nm. As interatomic potential functions, we used the all-atom OPLS-AA (Optimized Potentials for Liquid Simulations) force field for proteins and TIP3P (Transferable Intermolecular Potential with 3 Points) for water. To rule out force field-dependent effects, we confirmed our observations in tests applying also AMBER (Assisted Model Building with Energy Refinement), an alternative force field, which yielded similar results. Simulations were carried out using the GROMACS simulation package with a time step of 2 fs. For each run, 1,344 cores were used with a performance of about 85 ns/day. The size of the input file used in our simulations was 49 MB. Each simulation run produced 8 files. Except for the trajectory file, the sizes of the files were < 500 MB each. In total about 23 million core hours were used together with 5 TB of storage space reserved for the project.

The IR ectodomain system was successfully modelled and equilibrated in its inactive state (Fig. 2) and thus proved applicable for further studies that include insulin. Insulin exhibits complex binding characteristics featuring multivalent binding to the receptor. The hormone as well as each IR monomer contains two distinct binding surfaces. Insulin is envisioned to bind its receptor at one site first and then – with its second binding site – to bind additionally to the second binding site in the opposing receptor monomer thus establishing a cross-link. The precise mechanism of insulin binding, i.e., where and how it engages its receptor first, the process of cross-linking the receptor halves, and the dynamics of the structural transition have remained elusive to date.

Insulin was docked to the receptor ectodomain in its inactive state based on previous crystallographic data (PDB ID: 4XZB and 4OGA) [2]. It should be emphasized that the insulin/IR-ECD complex simulated here represents presumably only the initial state occurring during the activation process. This complex was found to maintain stable interactions throughout the simulation periods without insulin dissociating from the binding site. Nevertheless, we observed only minimal changes within the structure of the IR-ECD.

Conflicting models describing the mechanism underlying IR activation have been proposed. Very recently, we successfully reconstituted full-length insulin receptors in lipid nanodiscs, i.e., small artificial membrane patches, and directly visualized the conformational change in the receptor by single-particle electron microscopy (EM) [1, 3]. This insulin-induced structural change is complex and requires domain rearrangements as illustrated in Fig. 3. The insulin-bound IR-ECD conformation obtained further support by cryo-electron microscopy [4]. These recent structures enabled us to update our systems accordingly, but – due to the large extent of structural rearrangements in the receptor – call for longer simulation runs in order to reveal the complete transition.

Ongoing Research / Outlook

Combining electron microscopy data [3,4] with our simulation efforts will provide further insights into this activation mechanism. Currently, we are able to model insulin-free and insulin-bound IR-ECD complexes, equilibrate their structures, and consider the dynamics of those complexes in the given states, but further simulation studies are required to understand the complete transition into the activated state leading to downstream signaling (Fig. 3).

Simulating systems of large size remains a challenging task and was here possible only because of the significant resources granted by SuperMUC. Before running the production runs, the scaling of the simulation models was carried out on both Phase 1 and Phase 2 clusters. The scaling performance improved with Phase 2, thus speeding up the calculations. In production runs, the dynamics of the receptor appeared to be too slow to monitor large structural changes. Thus, simulations at longer time scales (> 20 microseconds) are required to observe the enormous – yet slow – changes reported in experimental studies [3, 4]. Simulations at extended time scales to simulate the activation process are ongoing. Deciphering the complete process of activation – resolved in space and time – remains of eminent importance to understand insulin action and to develop targeted strategies for treating pathologies such as diabetes.