Ramping activity in the medial prefrontal cortex integrates time and rewards in opposing directions during patch foraging

Abstract

As animals forage for food and water, a key decision they need to make is when to abandon a depleting resource to search elsewhere in the environment. The Marginal Value Theorem (MVT) provides a normative framework for this behavior: Animals should leave a depleting 'patch' of resources once the instantaneous reward rate in the patch drops below the average value of the environment. Computing the exact patch leaving time in this manner requires comprehensive knowledge of one’s environment – knowledge that animals do not have under naturalistic conditions. Despite this epistemic limitation, many species appear to well-approximate the MVT. How they do, however, remains largely unknown. To address this question, we designed a patch foraging task wherein head-fixed mice run along a 1-dimensional virtual track in search of patches containing water rewards. To mimic naturalistic conditions, rewards were delivered with varying probabilities, with probability decaying over time within patches to simulate a depleting resource. We confirmed a key prediction of the theory: Mice had greater patch residence times when rewards were larger or more frequent (p.0001 for both reward size and frequency, generalized linear mixed effects model, n = 18 mice). To investigate the neural basis of patch leave decisions, we recorded activity of neurons in the medial prefrontal cortex (mPFC) using high-density probes (Neuropixels) during the foraging task (3,683 neurons, 5 mice). We observed task-relevant dynamics throughout mPFC, most prominently in a neural population whose activity, in the absence of reward, ramped up over time as a patch leave decision became more imminent. When rewards were delivered, this same ramping population’s activity dipped, suggesting an evidence accumulation process that approximates the normative strategy by integrating time and rewards in opposing directions relative to a leave decision. Moreover, we confirmed that this pattern of ramping and reward suppression was prevalent on individual trials. Taken together, our results suggest an evidence accumulation process occurs during foraging decisions, and that it is widely represented as ramping activity throughout mPFC.