Inter-animal variability in learning depends on transfer of pre-task experience via the hippocampus

Learning occurs in the context of individual prior experience. Variability in learning is, therefore, due to both differences in learning capacity as well as differences in those prior experiences. To describe the effect of prior experience on learning and to identify how the brain transfers this prior experience for current learning, we measured the behavior of many rats across two tasks on a six-arm track. In the first, reward was delivered after any transition between arms. In the second, reward was only delivered at three arms (2-3-4). To receive reward, it was necessary to alternate between the outer arms after each visit to the center arm (e.g., 3-4-3-2). During the first task, rats displayed variability in the structure of their arm visits. The behavior ranged between sweeping across the track (e.g., 1-2-3-4-5-6-5-4-3-2-1), and only going back and forth between two arms. The propensity of the rats to sweep in the first task predicted almost a quarter of specific errors made while learning the second task, illustrating the effect of initial conditions on learning. In a separate cohort of rats, we found that lesioning the hippocampus disrupted this relationship, indicating that the hippocampus plays a role in this information transfer. To identify the computational principles underlying this hippocampus function, we applied a recently developed computational model: the Tolman-Eichenbaum Machine (TEM). TEM posits that hippocampal encoding results from extracting predictable structure from sequences of stimuli. We reasoned and confirmed that the behavioral variability measured in our first task would lead to differences in hippocampal encoding in TEM. Therefore, we find that naturally occurring behavioral variability is enough to generate distinct hippocampal codes, and that these differences could provide the substrate for a neural implementation of transfer learning.