Learning temporal structure engages hippocampus and guides value-based behaviour

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

This article is not in any list yet, why not save it to one of your lists.
Log in to save this article

Abstract

The ability to learn and exploit structured relationships between events is fundamental to adaptive behaviour and episodic-like memory, yet the neural mechanisms that support such learning remain poorly understood. Progress has been limited by the difficulty of dissociating relational structure from sensory cues, value, and motor output in experimentally tractable tasks. Here we introduce a temporally structured olfactory task for mice that isolates relational structure from cue identity and reward statistics. Mice rapidly learned to use the ordered relationship between two sequentially presented odours to predict outcome. Because individual odours were equally associated with reward across contexts, their predictive value depended entirely on their relationship to preceding cues. Behavioural analyses and reinforcement learning models revealed a gradual shift to the use of structured representations as learning progressed. Dopamine activity in the nucleus accumbens reflected this transition. Over learning, identical sensory inputs became associated with opposite predicted outcomes depending only on their temporal position, consistent with dopamine prediction errors reflecting latent task structure rather than cue identity. Trial-by-trial variability in dopamine dynamics was quantitatively captured by the same models that described behavioural strategy, linking dopamine prediction errors directly to these learned structural representations. Finally, using a matched control task that preserved sensory input and reward statistics while eliminating structure, we found that learning and using temporal structure selectively increased ventral hippocampal activity. Together, these results establish a minimal, non-spatial paradigm for studying structural learning in mice, link behavioural representations to dopamine signalling, and show that isolating relational structure from sensory and reward confounds selectively recruits hippocampal circuitry long theorised to support it.

Article activity feed