Computational protocol: Membrane protein orientation and refinement using a knowledge-based statistical potential

Similar protocols

Protocol publication

[…] The use of detergents for membrane solubilisation during transmembrane protein crystallisation means that information regarding the positions of lipid molecules in crystallographic data is extremely rare, making comparison with experimentally determined tilt angles difficult. A small number of transmembrane proteins have had their tilt angles determined using Attenuated Total Reflectance Fourier Transform Infrared (ATR-FTIR) spectroscopy. We assessed performance of the potential, using GA search, and 4 other methods – OPM, TMDET, Ez-3D and a potential derived from experimental measurements of free energy of membrane insertion described by Hessa et al. [] combined with a grid search - with these structures, comparing the mean absolute tilt angle of all transmembrane segments with the experimentally determined values (Table ). Hessa et al. used systematically designed hydrophobic segments to quantitatively analyse the position-dependent contribution of all 20 amino acids to membrane insertion efficiency. Results show that in all cases, tilt angles calculated by our potential correlate well with ATR-FTIR values. Given the large experimental error, we suspect that all five methods produce more or less equivalent estimates. In most cases the experimentally determined values are systematically larger, possibly due to orientational disorder under experimental conditions, suggesting that these experimental values represent the upper limits of the actual tilt angles [].We also tested the potential against the recently crystallised proton-gated urea channel HpUreI from Helicobacter pylori, a structure consisting of six protomers assembled in a hexameric ring surrounding a central bilayer plug of ordered lipids []. Applying the potential and GA to the unaligned structure, it was possible to position the channel such that the lipid vector average, formed by the vectors connecting the terminal carbon atoms in each of the lipid molecules in the cytoplasmic leaflet, was tilted from the z-axis by only 1.54 degrees. We also applied the potential in combination with a slow exhaustive search of all possible orientations, resulting in a lowest energy orientation where the lipid vector average was exactly parallel to the z-axis. [...] Table  summarises the performance of the combinatorial refinement algorithm incorporating the membrane potential when tested on 28 models generated by FILM3, showing TM-scores calculated over all helical Cα residues. The TM-score is intended to be a more accurate measure of structural alignment compared to RMSD or GDT. Scores are in the range (0,1], with 1 indicating a perfect match between two structures, scores below 0.20 typically correspond to randomly chosen unrelated proteins, while scores >0.5 are roughly the same fold []. Ten different weights were used for the membrane potential term, with a value of 1.6 producing the most consistent results. Compared to models generated using the standard combinatorial refinement procedure (column 2), models generated with the membrane potential (column 3) show an improved TM-score (> = 0.01) in 18 cases, with an average improvement across these 18 targets of 0.05. While increases in TM-score were generally modest, eight targets have TM-score increase of over 0.06, while two are over 0.1 (PDB IDs 2nq2A and 3dhwA). Only three targets have lower TM-scores after refinement with a decrease of 0.03 in the worst case, while seven targets remain unchanged. Across all 28 targets, the average TM-score change is 0.03. We also performed a second round of refinement using MODELLER following refinement using the membrane potential, comparing the resulting models to the final FILM3 models which had also been refined using MODELLER (columns 5–7). Results are similar with 16 targets improved, 6 unchanged and 6 made worse, again by only 0.03 in the worst case, and an average TM-score change is 0.03. These results indicate that different aspects of the structure are refined by the membrane potential and by MODELLER, suggesting that using both in combination should produce the best quality models. Across all Cα residues, performance is slightly less pronounced with 16 targets improved and 3 made worse, with an average TM-score change of 0.02. In terms of the positional accuracy lost by reducing the GA pool size, the maximum observed tilt error with a pool size of 500 was typically double that observed with a pool size of 10000. […]

Pipeline specifications