BEACON Seminar: Re-organisation of prefrontal cortical dynamics by decision making and across learning

The medial prefrontal cortex (mPFC) is a core contributor to flexible decision-making. Rodent mPFC can be anatomically subdivided into cingulate (ACC), prelimbic (PrL) and infralimbic (IL) cortex, yet the nature of information flow across these subdomains remains poorly defined. I will present some new experimental findings in which we addressed this using chronic extracellular recordings from the mPFC of adult rats implanted with Neuropixel silicon probes, permitting simultaneous capture of spikes and local field potential (LFP) from hundreds of neurons spanning ACC/PrL/IL cortices across over many months of recording.

Following implantation, rats were trained over 14 days to navigate a 3-armed maze using a cyclic alternation rule for sucrose reward, allowing us to quantify the encoding of behavioral variables by task context-dependent firing patterns of individual neurons in both ‘naïve’ and ‘expert’ rats. Firing rate modulation of mPFC single neurons provided rich information about start and end locations of the animals’ runs, the readout of which peaked at choice points, consistent with a role of these neurons in decision-making. We observed a robust increase in neurons' mixed retrospective and prospective coding of start/goal locations at choice points over the course of learning. We did not find any evidence for anatomical segregation of the encoding of task variables or reward anticipation/response between ACC/IL/PrL.  Moreover, we detected cell ensembles between neurons separated by up to 5mm of tissue defined on the basis of stereotyped spike patterns at occurring at 5-500ms resolution. Cell ensemble activities, organized by phase of the LFP theta (5-10Hz) and gamma (30-80Hz) -band cycles, were observed between ~20% of potential pairwise unit interactions. Compared to constituent single units, assembly activation patterns provided stronger readouts of behavioral variables. Multi-neuron spike patterns of cell ensembles could link neurons with disjoint rate-encoded variables to represent different task-related information, thereby supporting a multiplexed spike rate-time code. We also observed changes in the spatio-temporal and cell-type specific logic of assembly interaction over the course of task learning.

This work demonstrates both integration and segregation of functional information processing in the mPFC and provides novel insights into the logic and experience-dependence of network computations supporting decisions.