Conformational landscape of 2-aminopurine-substituted RNA oligonucleotides from machine-learning-driven enhanced sampling
Fluorescent base analogs, such as 2-aminopurine (2AP), are frequently used to investigate nucleic acid conformational dynamics, particularly base-stacking interactions, through fluorescence spectroscopy. Although 2AP substitution can induce structural perturbations, properly accounting for these effects enables its use as a probe of complex RNA dynamics. Here, we apply explainable machine-learning-derived surrogate-model collective variables to perform enhanced-sampling simulations that comprehensively explore the free-energy landscapes of both 2AP-substituted and unsubstituted RNA dinucleotides and trinucleotides. Our results show that while 2AP substitution does not significantly alter stacking propensity, it substantially modulates the free energy landscape, leading to a redistribution of populations among stacking modes. The fraction of stacked 2AP conformations closely correlates with experimentally observed “dark state” populations, supporting the hypothesis that base stacking quenches 2AP fluorescence and promotes dark state formation. This work and follow-up studies in larger, physiologically relevant RNA systems will establish the ability to correlate observed emissive or dark states of 2AP with specific structures within the free-energy landscape.