Excessive aggregation of membrane proteins in the Martini model

Abstract

The coarse-grained Martini model is employed extensively to study membrane protein oligomerization. While this approach is exceptionally promising given its computational efficiency, it is alarming that a significant fraction of these studies demonstrate unrealistic protein clusters, whose formation is essentially an irreversible process. This suggests that the protein-protein interactions are exaggerated in the Martini model. If this held true, then it would limit the applicability of Martini to study multi-protein complexes, as the rapidly clustering proteins would not be able to properly sample the correct dimerization conformations. In this work we first demonstrate the excessive protein aggregation by comparing the dimerization free energies of helical transmembrane peptides obtained with the Martini model to those determined from FRET experiments. Second, we show that the predictions provided by the Martini model for the structures of transmembrane domain dimers are in poor agreement with the corresponding structures resolved using NMR. Next, we demonstrate that the first issue can be overcome by slightly scaling down the Martini protein-protein interactions in a manner, which does not interfere with the other Martini interaction parameters. By preventing excessive, irreversible, and non-selective aggregation of membrane proteins, this approach renders the consideration of lateral dynamics and protein-lipid interactions in crowded membranes by the Martini model more realistic. However, this adjusted model does not lead to an improvement in the predicted dimer structures. This implicates that the poor agreement between the Martini model and NMR structures cannot be cured by simply uniformly reducing the interactions between all protein beads. Instead, a careful amino-acid specific adjustment of the protein-protein interactions is likely required.

Fig 1. Selected snapshots of transmembrane domain dimerization.

Examples of conformations at different COM–COM distances (from top to the bottom: 1.2, 2.0, and 2.4 nm) from the umbrella sampling simulations (COM stands for the center of mass). Here, the U-0.1 scaling method is employed (see below). Chains of lipids (DLPC) are shown in cyan, phosphate beads in brown, and choline beads in blue. Peptides (EphA1) are shown in yellow and orange. Water, antifreeze particles, and ions have been omitted from the images.

Fig 2. Dimerization free energy profiles of transmembrane domains.

Free energy profiles of EphA1 and ErbB1 TM domain dimerization using either unmodified Martini or the “U” and “W” scaling strategies. The solid gray line at a value of zero is to guide the eye. The dashed gray line shows the experimental value for ΔG_DIM. Note that the common legend is split between the two panels.

Fig 3. Deviation of the spontaneously formed dimer structures from their NMR structures.

Data is shown for both the standard (non-polarizable) Martini and the U-0.1 scaling. The root-mean-square deviation (RMSD) is shown on the y axis, and the absolute value of the deviation of the dimer crossing angle on the x axis. The coloring shows the deviation in the number of backbone bead contacts (cutoff of 0.8 nm). Therefore, an optimal result would be a purple dot (correct number of contacts) at the origin (RMSD equal to 0 nm and with the correct dimer crossing angle). As a rule of thumb, to guide the eye, the acceptable region (deviations up to 10° and 1 nm considered to be acceptable) is highlighted in green. In each graph, data are shown for 10 replica simulations. Each replica is depicted with a marker with error bars showing standard deviation, while the mean over replicas is highlighted by the dashed lines. The crossing angle of the NMR structure is given for each dimer in degrees. Data are extracted from the last 20 μs of the simulations.

Fig 4. Final structures of the 10 validation replica simulations.

Shown here are the structures for both standard (non-polarizable) Martini and the best scaling strategy resolved in this work (U-0.1). The two peptides in the NMR structure are shown in yellow (first peptide) and green (second peptide). The dimerization motif (see Ref. [71]) of the first peptide is additionally highlighted in purple. The dimer structures based on the simulations are depicted in a manner, where the first peptides of the replicas (all shown in gray) are RMSD-aligned at the dimerization motif (purple) on the first peptide of the NMR structure (yellow). The reader should here concentrate on the second peptide structures (not aligned) given by the 10 replica simulations that are depicted by a spectrum of red-white-blue. If there is good agreement between simulation and NMR, the structures of the second peptide (red-white-blue) should coincide with the second peptide in the NMR structure (green).