Biological applications of discrete molecular dynamics

Abstract

Sequence, structure and dynamics are an indivisible tandem to understand protein function. Luckily, evolution imposed a hierarchical rational between that facilitates the analysis: dynamics are encoded in the structure, which in turn, is encoded in the sequence. Decipher the mechanisms governing protein function requires contributions from diverse fields, particularly to follow molecular motions. There are technological limitations to monitor local, elemental, protein movements, since they are too fast to be followed by current experimental set-ups. Theoretical models provide necessary assistance in this regard mainly through molecular simulations. But atomistically simulations of large functional motions make computations, currently, unaffordable. The problem is that large-scale motions are rooted in the very fast elemental ones; so, in order to observe a biological-functional conformational change we have to keep track of all the elemental motions occurring. The gap in the time scale of both extremes of motions is devastating: fast motions are over 1015 times faster than functional ones.

In this Thesis, I present our contribution to extend the simulation time range, in an effort towards more predictive computational models. We explored alternative methods to retrieve molecular motions from the underlying physical forces governing proteins. The method used is named Discrete Molecular Dynamics and represents by itself a significant improvement in computational efficiency. In order to go further, we lower the resolution of protein models to a coarse-grained representation both in terms of number of particles and interaction functions. We benefited from several existing algorithms to simplify calculations keeping the models as much accurate as possible. Putting all this methodological innovations together, we developed models to follow conformational transitions of proteins, from local re-arrangements to motions changing drastically the protein structure. Also, we applied novel computational approaches to account for protein flexibility upon recognizing and binding other interacting proteins.

In a second stage, we investigated the echo of protein flexibility and dynamics printed out in the sequence of the protein. We observed over the history of the sequence that instead of one single native structure, proteins were tuned to have several conformations. We exploited this flexibility signature in the sequence to predict protein motions and eventually alternative protein conformations. Finally, we use our efficient tools to move protein dynamics analysis to the proteome scale. We searched for all proteins having two known conformations, a symptom of a conformational transition, and then, we used those conformations to follow the motion from one state to the other. We analyzed and structured all that dynamical information of proteins and connected our results to the most detailed simulation methods available to dissect the fine details of proteins dynamical behavior when required.

Secuencia, estructura y dinámica forman un trío un insoslayable en el funcionamiento de las proteínas. El proceso evolutivo codificó la dinámica en la estructura de las proteínas, que a su vez, está codificada en la secuencia. Descifrar los mecanismos que rigen el movimiento de las proteínas requiere la fusión de experimentos y modelos teóricos. Los modelos teóricos proporcionan asistencia necesaria a través de simulaciones moleculares, pero su costo computacional es tan elevado que puede impedir el estudio. El problema radica en que los movimientos biológicamente interesantes son la consecuencia de un cúmulo de movimientos de alta frecuencia, que es necesario seguir para comprender los movimientos funcionales. La brecha entre ambos tiempos asciende a un impresionante ratio de 1015.

En esta Tesis, presento métodos para aumentar la eficacia de los cálculos moleculares con el objetivo de acortar la diferencia entre el tiempo de lo que es simulable a lo que es biológicamente interesante. El método utilizado es Discrete Molecular Dynarnics y representa por sí mismo una mejora significativa en la eficiencia computacional. En resumen, hemos desarrollado modelos para seguir transiciones conformacionales de proteínas, desde movimientos locales hasta otros que cambian radicalmente la forma de la proteína. Dichos métodos fueron aplicados tanto a transiciones conformacionales como a interacciones proteína-proteína. En una segunda etapa, buscamos la imprenta en la secuencia del patrón de flexibilidad de la proteína, con el objetivo de predecir los cambios de conformación. Finalmente, utilizando los métodos desarrollados hemos concluido un análisis a gran escala sobre la dinámica de las proteínas, simulando todas las transiciones cuyos dos extremos fueron determinados experimentalmente. Los resultados de dichas simulaciones fueron integrados con los métodos de simulación más fiables disponibles, para aumentar en nivel de detalle cuando sea necesario.