DOPAMINE PREDICTION ERRORS IN REWARD-FREE STATISTICAL LEARNING

Statistical learning (SL) allows organisms to infer latent structure from sensory input without instruction or reinforcement, a process essential for building internal models of the environment. We recently demonstrated that the hippocampus, specifically dorsal CA1 (dCA1), is causally required for this capacity. Using a novel reward-independent auditory task, we showed that mice spontaneously acquire abstract structural rules, such as event frequency, and sequence abstract structure, tracked by pupil dilation as a physiological readout of expectation violation. High-density recordings revealed that dCA1 supports this learning by dynamically reorganizing population activity into orthogonal subspaces that separately encode sensory features and abstract contexts.
However, the upstream mechanisms driving this hippocampal reorganization remain unknown. While ventral tegmental area (VTA) dopamine is traditionally linked to reward prediction errors, growing evidence suggests it also tracks value-neutral sensory surprise, making them a candidate mechanism for driving updates in hippocampal models. We propose that the VTA provides a non-reward prediction error signal that triggers the "shift" in hippocampal neural subspaces observed during learning. To test this, we are recording VTA dopamine dynamics using fiber photometry during our SL paradigm. Furthermore, we are selectively inactivating the VTA-to-hippocampus pathway to determine if dopaminergic input is necessary for updating hippocampal internal models. This work aims to mechanistically link midbrain error signals to the formation of abstract cognitive maps in the absence of reward.