Development and assessment of methods for membrane protein structure prediction

今年，我们在两个重要目标上取得了进展：提高膜蛋白对称性检测的准确性，以及开发膜蛋白集合的定量描述，如下所述。

This year we made progress on two important goals: improving the accuracy of detection of symmetry in membrane proteins, and developing quantitative descriptions of membrane proten ensembles, as described below. Membrane proteins contain abundant amounts of structural symmetry and those symmetries can provide the foundation of the function of that protein. Several computational methods are available to detect those relationships, but their accuracy for membrane proteins has not been systematically tested. A first step toward such an assessment is the development of a comprehensive benchmark against which methods can be tested. This year, we completed development of the MemSTATS set of symmetry-containing proteins. We tested available methods on this benchmark, and identified a large number of cases in which symmetries are not detected (Ref. 1). This study identified opportunities for development of improved and automated methodologies for detecting membrane protein symmetries that can be related to their functional mechanisms and evolutionary origins. The benchmark will provide an important foundation for further development of our database called EncoMPASS (Encyclopedia for Membrane Proteins Analyzed by Structure and Symmetry), which, with the assistance of NINDS intramural Bioinformatics staff (Yavaktar and Kumar), is available through a webserver hosted at https://encompass.ninds.nih.gov. Membrane protein functional cycles often involve multiple and dramatic changes in conformation, for example, in response to binding of ligands. Moreover, each of those individual states is actually best reflected by an ensemble of conformations. Mapping out those conformational ensembles is, however, extremely challenging. In principal, biophysical methods that examine proteins in solution, such as electron paramagnetic resonance or hydrogen-deuterium exchange can provide this level of information, and can be layered on top of the static structures provided by structural biology studies. In practice, however, translating the data from these biophysical methods lacks rigor and can be challenging to interpret. To this end, we developed an approach designed to integrate data from hydrogen-deuterium exchange data with structural models obtained from, e.g., molecular dynamics simulations. This method, called HDXer, was developed together with the Faraldo-Gomez lab at NHLBI and is based on maximum entropy theory (Ref. 2), enabling it to quantify the extent to which the conformations of the structure match the biophysical data. We illustrated the applicability of the method first with artificial data generated for a water-soluble periplasmic binding protein, and secondly with real-life experimental data obtained for a membrane transport protein, LeuT. In both cases, HDXer was shown to distinguish ensembles found in a large conformational change. HDXer may therefore prove to be an important tool to leverage a plethora of experimental data in more quantitative ways.