Circular RNAs may embed pieces of real-world sensory information into an episodic memory

For a generation, neuroscience has searched for a molecule that stores our memories across time. This search has focused on proteomic mechanisms, but less is known about RNA. Here, we identify a new persisting class of RNA associated with long-term memory - Circular RNAs. Unlike other RNAs, Circular RNAs are stable for days or longer and may provide a means for storing sensory information across time. We leveraged a differential fear conditioning paradigm whereby individual mice sample all real-world sensory inputs (i.e., auditory, visual, gustatory, olfactory, and incidental tactile) in a quasi-stochastic manner prior to receiving different intensities of an unconditioned stimulus (US) foot-shock. While Pavlovian models of learning from the 20th century were critical for understanding elemental associations, they fail to appreciate (1) what US content remains inside of a complex conditioned stimulus (CS) or response (CR - a behavioral manifestation of an episodic memory), (2) what happens when the associations involve multiple senses, and (3) what biologically happens to the real-world US. Given (1) we are constantly sampling information from our environment through all our senses and (2) the US at a given moment in time likely adds value to imprint that multisensory representation, we propose the real-world US is biologically encoded via back-spliced Circular RNAs within the cells and circuits that represent a particular episodic memory and present days later. This logic, best simplified by the equation: [Formula], allowed us to ask how the formation of similar episodic memories, which only differ in relation to the content of US information, alter Circular RNAs in the CA1 subfield of the hippocampus - a brain area critical for episodic memories. We found that stronger foot-shock USs during conditioning produce stronger memories relative to weaker USs 24-h later. Stronger memories also generalize to novel/safe environments 48-h later. Moreover, the unconditioned response is highly correlated with future CRs, suggesting (1) an understudied relationship between the strength and type of US/URs and future CRs in complex environments as well as (2) fear generalization, at least in the short-term, is associated with the embedding of additional US information. Next-generation Circular RNA sequencing 1-hr after acquisition revealed a remarkably small set of circular RNAs relative to nearly identical, yet weaker, episodic memories in CA1. Gene Ontologies for mice that formed weaker and stronger memories matched those families classically involved in weaker and stronger forms of memory across species. Preliminary in situ hybridization visually confirmed the presence of Circular RNAs in the CA1 subfield. Future experiments will examine the persistence of Circular RNAs in cells of a memory trace (i.e., engram cells; in situ hybridization) at recent (4 days) and remote (21-days) time points. Taken together with our mathematical model for multisensory learning, our data suggest that Circular RNAs do not contribute to the storage of the multisensory configural representation, but perhaps to the storage of discrete pieces of real-world sensory information related to the US that is partially embedded inside of a memory trace early-on. Importantly, in the above model for multisensory learning, the discrete USs are biologically separable from the future CS(NS+US) associations and US strength is modifiable across time. This work reveals fundamental insights into how we store pieces of real-world sensory information in an episodic memory at the biological level of the brain. One Sentence SummarycircRNAs biologically encode real-world sensory information into a long-term memory