Simulation and analysis of FRET in the study of membrane proteins

Membrane proteins play an important role in almost all cell activities. However, the characterization of the structure of membrane proteins in lipid bilayers is still at the frontier of structural biology. While 30-40% of all proteins are situated at or in membranes, yet less than 1% of the known protein structures are of membrane proteins. The complexity and delicacy of membrane-protein systems impedes the application of standard methods of protein study, such as X-ray crystallography and NMR. Furthermore, these techniques are mainly aimed at short-range structural information, and seem to be not very useful for the study of long-range interactions, for instance in the case of protein assemblies. These factors urge to find other biophysical approaches to study proteins incorporated into lipid bilayers. A successful alternative is Förster (or fluorescence) resonance energy transfer (FRET) spectroscopy in combination with site-directed labeling with fluorescent probes. This technique provides distance information within a range of 10-100 Å, which is sufficient to study the structure of membrane proteins and their complexes. The crucial point in the extraction of structural information from FRET data is an advanced and robust data analysis. This work is devoted to the development of such methods for analysis of fluorescence data, based on simulation modeling, global analysis and artificial neural networks. Especially the advances and problems of the simulation-based fitting (SBF) approach to fluorescence data analysis are considered. The methodologies of global analysis and SBF are applied to obtain information about the position, aggregation and structure of M13 major coat protein in DOPC:DOPGvesicles. The resulting physical parameters, that describe the embedment and orientation of the protein in the membrane, such as protein-protein aggregation, protein depth, tilt angle, and tilt direction, are in good accordance with previously reported values. Based on the FRET data, it was found that M13 major coat protein (having 50 amino acid residues) in its bilayer conformation could be described as a singlea-helix between amino acid positions 10-46. Additional work was performed on the methodological aspects of improving the SBF data analysis technique. Here it is proposed to use an artificial neural network to speed up the parameter identification and to make the process of fitting less sensitive to noise. The main idea of this method is the substitution of a time-consuming simulation model by an artificial neural network, specifically a multi-layer perceptron. The method results in a speeding up of the simulation by about a factor of 10 ⁴ for the developed FRET model.