Complementary task representations in hippocampus and prefrontal cortex for generalising the structure of problems

Abstract

Few situations in life are completely novel. We effortlessly generalise prior knowledge to solve novel problems, abstracting common structure and mapping it onto new sensorimotor specifics. Here we trained mice on a series of reversal learning tasks that shared the same structure but had different physical implementations. Performance improved across tasks, demonstrating transfer of knowledge. Neurons in medial prefrontal cortex (mPFC) maintained similar representations across multiple tasks, despite their different sensorimotor correlates, whereas hippocampal (dCA1) representations were more strongly influenced by the specifics of each task. Critically, this was true both for representations of the events that comprised each trial, and those that integrated choices and outcomes over multiple trials to guide subjects' decisions. These data suggest that PFC and hippocampus play complementary roles in generalisation of knowledge, with the former abstracting the common structure among related tasks, and the latter mapping this structure onto the specifics of the current situation.

Dopamine-independent state inference mediates expert reward guided decision making

“…These data build on previous evidence for inference underlying behavioural flexibility [12][13][14][15][16]18 , and informing dopamine responses 15,22,24,27,40 , but extend them by both recording and manipulating dopamine during inference-guided adaptive decisions, revealing a disconnect between value reporting and policy updating. An inference-based perspective helps explain previous observations that are surprising under a pure value learning account of flexible behaviour.…”

“…Specifically, model-free values learnt conditioned on the latent state will be high for the first-step action that commonly leads to the second-step port with high reward probability. These data build on previous evidence for inference underlying behavioural flexibility 12,13,15 , and informing dopamine responses 18,20,34,35 , but extend them by both recording and manipulating dopamine on fast timescales during inference-guided adaptive decisions, revealing a disconnect between value reporting and policy updating. An inference-based perspective helps explain previous observations that are surprising under a pure value learning account of flexible behaviour.…”

“…Such structured change is also common in the laboratory, and state inference could in principle underlie behavioural adaptation in many reward-guided tasks 10,11 . Indeed, direct evidence for inference based behavioural flexibility has been demonstrated in several tasks [12][13][14][15][16] , and is also suggested by meta-learning, where subjects get faster at adapting to changes with repeated experience [17][18][19][20][21] . Therefore, a pressing issue is to understand how dopamine reflects and shapes behaviour during inference-guided choice.…”

The generalized spatial representation in the prefrontal cortex is inherited from the hippocampus

“…Altogether, this corroborates the idea that hippocampal activity is also important for egocentric coding (65). We also expanded upon our understanding of the generalization of mPFC spatial representation (24)(25)(26), and its extent, by comparing the trajectory-specificity of the maps and their effect on the decoding (Fig. 1).…”

“…Previous reports characterized the spatial information conveyed by medial prefrontal neurons, and showed that both single neurons and population activity provide a generalized spatial code (24)(25)(26). Here, we tested whether generalized representations in the medial prefrontal cortex result from a bidirectional interaction with, or are mainly derived from spatial representations already present in the hippocampus.…”