Where Actions Meet Outcomes: Medial Prefrontal Cortex, Central Thalamus, and the Basal Ganglia

Abstract

Medial prefrontal cortex (mPFC) interacts with distributed networks that give rise to goal-directed behavior through afferent and efferent connections with multiple thalamic nuclei and recurrent basal ganglia-thalamocortical circuits. Recent studies have revealed individual roles for different thalamic nuclei: mediodorsal (MD) regulation of signaling properties in mPFC neurons, intralaminar control of cortico-basal ganglia networks, ventral medial facilitation of integrative motor function, and hippocampal functions supported by ventral midline and anterior nuclei. Large scale mapping studies have identified functionally distinct cortico-basal ganglia-thalamocortical subnetworks that provide a structural basis for understanding information processing and functional heterogeneity within the basal ganglia. Behavioral analyses comparing functional deficits produced by lesions or inactivation of specific thalamic nuclei or subregions of mPFC or the basal ganglia have elucidated the interdependent roles of these areas in adaptive goal-directed behavior. Electrophysiological recordings of mPFC neurons in rats performing delayed non-matching-to position (DNMTP) and other complex decision making tasks have revealed populations of neurons with activity related to actions and outcomes that underlie these behaviors. These include responses related to motor preparation, instrumental actions, movement, anticipation and delivery of action outcomes, memory delay, and spatial context. Comparison of results for mPFC, MD, and ventral pallidum (VP) suggest critical roles for mPFC in prospective processes that precede actions, MD for reinforcing task-relevant responses in mPFC, and VP for providing feedback about action outcomes. Synthesis of electrophysiological and behavioral results indicates that different networks connecting mPFC with thalamus and the basal ganglia are organized to support distinct functions that allow organisms to act efficiently to obtain intended outcomes.

Keywords: action outcome contingency; adaptive decision making; anterior cingulate cortex; intralaminar nuclei; mediodorsal nucleus; prefrontal cortex; reward guided choice; ventral pallidum.

FIGURE 1

Normalized population histograms for response types observed in mPFC for rats performing the dynamic DNMTP task. The task is illustrated above. Trials begin with a randomly selected lever extending at one of four possible locations (90° apart) for the start response. The start lever retracted when pressed and a lever was then extended 90° to the left or right (randomly selected) of the start lever. This retracted when pressed and water reinforcement delivered (indicated by asterisk) through a spout immediately above the lever. The initial lever was then reinserted for the delay response. This was retracted with the first press after the memory delay ended and levers 90° to the left and right inserted for the choice. These both were retracted when either one was pressed and reinforcement delivered when the lever not extended for the sample was the one pressed (a correct non-matching response). Results are shown for all neuronal responses recorded for each response type with a minimum of 40 trials completed in a 60 m session. These were averaged for individual neurons and normalized so that each response recorded contributed equally to the population function. Error bars represent standard error of the mean. Results are shown for preparation (n = 44), movement 1 (before all lever presses, n = 97), movement 2 (toward reinforced lever presses, n = 32), lever press (n = 28), base lever press (start and delay presses only, n = 30), reinforcement anticipation (preceding delivery, n = 50), reinforcement (following delivery, n = 63), delay (n = 58), post-reinforcement (when rats disengaged from spouts, n = 16), and error (n = 4). These population histograms were previously published online (Francoeur and Mair, 2018). ? Represents choice.

FIGURE 2

Schematic summarizing the main connections of medial prefrontal cortex (mPFC), including interconnections with striatum and central thalamus. Pathways are color coded to identify afferent and efferent connections of the main subregions of mPFC: secondary motor (M2), anterior cingulate (AC), prelimbic (PL), and infralimbic (IL) cortices. The weight of lines indicates heavy, moderate, or light projections. Unidirectional or bidirectional transmission is indicated by arrowheads. Projections are shown for intralaminar (ILn), ventromedial (VM), mediodorsal (MD), midline (Mid), and anterior medial (AM) nuclei in central thalamus. The division of cortex into limbic and non-limbic regions follows Hoover and Vertes (2007). Estimates of mPFC projection densities rely primarily on Vertes (2002, and Hoover and Vertes (2007). See text for details.

FIGURE 3

Normalized population histograms responses related to reinforcement delivery in medial prefrontal cortex (mPFC; n = 63), the mediodorsal thalamic nucleus (MD, n = 71), and ventral pallidum (VP, n = 101) and reinforcement anticipation in mPFC (n = 50) and MD (n = 46). Anticipatory responses were not observed in VP. Anticipatory responses began on average 0.8 s before rewards were normally delivered and ended 0.2 s after reward delivery ended or 0.2 s errors when rewards were not delivered. Reinforcement responses began 0.2 s after reward delivery began and lasted until 1.0 s after reward delivery ended. Results are shown for all neuronal responses recorded for each response type with a minimum of 40 trials completed in a 60 m session. These were averaged for individual neurons and normalized so that each response recorded contributed equally to the population function. Error bars represent standard error of the mean.

FIGURE 4

Effects of mPFC, hippocampal, intralaminar thalamic (ILn), and striatal lesions on varying- and recurring-choice DNMTP trained in automated eight arm radial mazes. In varying choice DNMTP arms were selected at random from all available options for start, sample, delay, and choice responses in a lighted room with many visible external cues to favor an allocentric solution and to eliminate the possibility that a correct choice could be determined prospectively. In recurring choice, the same three arms (in a T-configuration) were used on every trial for start/delay and left and right sample/choice responses and mazes were covered and the room darkened to minimize external cues and favor egocentric choice and prospective decision making. mPFC lesions spared varying-choice and produced delay-independent deficits for recurring-choice DNMTP. Hippocampal lesions produced delay-dependent deficits for both versions, consistent with rapid decay of working memory. ILn lesions produced delay-independent deficits for both versions, consistent with the effects of dorsomedial striatal lesions. The ventral striatal lesion group were impaired for recurring-choice DNMTP and were impaired compared to the dorsolateral but not the control group for varying-choice DNMTP. Error bars represent standard error of the mean.

FIGURE 5

Effects of frontal cortical, striatal and thalamic lesions on visuospatial reaction time (VSRT). Rats performed an observational (runway) response, traveling down a runway after pressing a lever to start the trial, so that they entered the test chamber facing an array of response ports. Entering the test chamber triggered a brief luminance in one of the ports in which water reward was delivered if rats made a nose poke in that port first within a 3.0 s limited hold. Choice accuracy is plotted as a function of stimulus duration (varied randomly from trial to trial). Response time (RT) is plotted for the observational runway response, performed without modification at the start of each trial, and for the stimulus guided choice response. A similar pattern of impairment was observed for lesions of M1M2 motor cortex, dorsolateral striatum, and intralaminar thalamic nuclei (IL): stimulus duration dependent impairment of choice accuracy coupled with an increase in choice, but not runway, RT.

FIGURE 6

Effects of mPFC, hippocampus, thalamic, striatal, and ventral pallidal lesions on DMTP trained in operant chambers with 3 retractable levers: two on the front wall for sample and choice responses and one on the back wall to initiate trials and to force rats to disengage from front wall levers during memory delays. Accuracy is plotted as percent correct as a function of memory delay and response time (RT) as cumulative functions showing the percentage of choice responses made in 1 s bins after the end of the memory delay. Hippocampal and mediodorsal thalamic (MD) lesions produced delay-dependent impairment of accuracy and had no effect on RT. Dorsolateral lesions had no significant effect on accuracy or RT. Dorsomedial and nucleus accumbens (Acb) core produced delay independent deficits in accuracy, significantly and had no significant effect on RT. mPFC, ventral pallidum, ventral striatum, and Intralaminar (IL), ventromedial (VM), and combined intralaminar/mediodorsal (ILMD) thalamic lesions produced delay-independent deficits in accuracy and increased median RT significantly.

FIGURE 7

The percentage of units recorded that meet the criteria for an isolated neuron that also meet criteria for specific response types (see Francoeur and Mair, 2018). Movement and lever press types are combined into single categories. Results are compared for ventral pallidum (VP), mediodorsal thalamus (MD), and medial prefrontal cortex (mPFC).

FIGURE 8

Normalized population histograms comparing responses of neurons with elevated activity during each lever press in mPFC (n = 28) and ventral pallidum (VP; n = 22) and neurons with elevated activity during movements toward each lever press for mPFC (n = 97) and mediodorsal thalamus (MD; n = 91). Results are shown for all neuronal responses recorded for each response type with a minimum of 40 trials completed in a 60 m session. These were averaged for individual neurons and normalized so that each response recorded contributed equally to the population function. Error bars represent standard error of the mean.