Leonardo Monrroy: Time-resolved X-ray Studies of Protein Structural Dynamics: Experimental and Computational Advances

Abstract

Understanding protein function requires insight not only into static structures but also into thedynamic processes that govern conformational changes over time. X-ray–based techniques, suchas time-resolved X-ray solution scattering (TRXSS) and time-resolved serial crystallography(TRSX), provide powerful means to probe these dynamics across timescales ranging fromfemtoseconds to milliseconds. However, the extraction of reliable structural informationremains challenging due to the low information content of the data, ensemble averaging effects,and the complexity of linking experimental signals to underlying molecular structures.This thesis aims to advance both experimental and computational strategies for studyingprotein structural dynamics. In Paper I, a novel data acquisition approach for TRXSS at MHzrepetition-rate X-ray free-electron lasers is introduced, enabling rapid and efficient collectionof time-resolved scattering data. This strategy enhances signal sensitivity and allows detectionof weak transient states that are otherwise difficult to resolve. Application to photoreceptorsystems reveals previously hidden intermediates and underscores the importance of efficientdata acquisition for capturing subtle structural changes.In Paper II, time-resolved crystallography is applied to investigate the photocycle of abathy phytochrome, providing atomic-level insight into ultrafast chromophore isomerizationand subsequent structural evolution. While the method resolves local structural changes withhigh precision, it also highlights limitations in capturing large-scale motions due to constraintsimposed by the crystal lattice.In Paper III, a machine learning–based framework is developed to address challengesin modeling TRXSS data. A variational autoencoder trained on electron densities derivedfrom molecular simulations and structure prediction methods enables the reconstruction ofconformational ensembles directly in real space. This approach captures structural heterogeneityand mitigates the non-uniqueness of the inverse scattering problem by implicitly inferringphysically meaningful, chemically consistent restraints on electron densities through the learnedlatent representation.A key theme throughout the thesis is the importance of accurately accounting for experimentalparameters, particularly the photoactivation yield, which critically links signal amplitude tostructural interpretation. Altogether, this work demonstrates how advances in experimentaldesign, data analysis, and machine learning can improve the interpretation of time-resolved X-ray experiments and provide a more comprehensive view of protein dynamics