Mapping energy landscapes of homopolymeric RNAs via simulated tempering and deep unsupervised learning

Abstract

Conformational dynamics plays crucial roles in RNA functions about sensing and responding to environmental signals. The liquid-liquid phase separation of RNAs and the formation of stress granules partly relies on RNAs conformational plasticity and its ability to engage in multivalent interactions. Recent experiments with homopolymeric and low-complexity RNAs have revealed significant differences in phase separations due to differences in base chemistry of RNA units. We hypothesize that differences in RNA phase-transition dynamics can be traced back to the differences in conformational dynamics of single RNA chains. In the present contribution, we utilize atomistic simulations with numerous unsupervised learning to map temperature dependence conformational free energy landscapes for homopolymeric RNA chains. These landscapes reveal a variety of metastable excited states influenced by the nature of base chemistry. We shed light on the distinct contributions of the polyphosphate backbone versus base chemistry in shaping conformational ensembles of different RNAs. We demonstrate that the experimentally observed temperature-driven shifts in metastable state populations align with experimental phase diagrams for homopolymeric RNAs. The work establishes a microscopic framework to reason about base-specific RNA propensity for phase separation. We believe our work will be valuable for designing novel RNA sensors for biological and synthetic applications.