Computational modeling and analysis of hippocampal-prefrontal information coding during a spatial decision-making task

Thomas Jahans-Price,Thomas E Gorochowski,Rafal Bogacz,Matthew W Jones,Matthew A Wilson

doi:10.3389/fnbeh.2014.00062

Abstract

We introduce a computational model describing rat behavior and the interactions of neural populations processing spatial and mnemonic information during a maze-based, decision-making task. The model integrates sensory input and implements working memory to inform decisions at a choice point, reproducing rat behavioral data and predicting the occurrence of turn- and memory-dependent activity in neuronal networks subserving task performance. We tested these model predictions using a new software toolbox (Maze Query Language, MQL) to analyse activity of medial prefrontal cortical (mPFC) and dorsal hippocampal (dCA1) neurons recorded from six adult rats during task performance. The firing rates of dCA1 neurons discriminated context (i.e., the direction of the previous turn), whilst a subset of mPFC neurons was selective for current turn direction or context, with some conjunctively encoding both. mPFC turn-selective neurons displayed a ramping of activity on approach to the decision turn and turn-selectivity in mPFC was significantly reduced during error trials. These analyses complement data from neurophysiological recordings in non-human primates indicating that firing rates of cortical neurons correlate with integration of sensory evidence used to inform decision-making.

Highlights

Single cell recordings from frontal and parietal cortical areas during simple choice tasks indicate that the activity levels of neurons in these areas integrate sensory evidence for the available alternatives over time (Kim and Shadlen, 1999; Schall, 2001; Shadlen and Newsome, 2001; Hanks et al, 2012); decisions are made when accumulated activity in these neurons reaches a threshold (Roitman and Shadlen, 2002)
Computational approaches like the leaky competing accumulator (LCA) model have been developed for perceptual choice tasks, but are not directly able to describe maze-based spatial tasks typically used in rodent studies where decisions about turn direction are based on information held in memory
COMPUTATIONAL MODEL Simulations were run to test if the computational model summarized in Figure 2A could generate similar behavior to that seen under experimental conditions

Summary

Introduction

Single cell recordings from frontal and parietal cortical areas during simple choice tasks indicate that the activity levels of neurons in these areas integrate sensory evidence for the available alternatives over time (Kim and Shadlen, 1999; Schall, 2001; Shadlen and Newsome, 2001; Hanks et al, 2012); decisions are made when accumulated activity in these neurons reaches a threshold (Roitman and Shadlen, 2002) This view of decision-making has been formalized in computational models that capture data describing both behavior and neural activity (e.g., Usher and McClelland, 2001; Wang, 2002; Mazurek et al, 2003; Ditterich, 2006; Beck et al, 2008). The extent to which diverse turn-, route- and memory-dependent firing correlates in rodent mPFC parallel decision-related activity in primate data remains unclear

Methods

Results

Discussion

Conclusion