Introduction

Working memory functions subserved by the frontal lobes are highly dependent on mesocortical dopamine (DA) transmission (Williams and Goldman-Rakic, 1995; Robbins, 2000). Electrophysiological recordings in freely behaving rats and primates have shown that neural activity in the prefrontal cortex (PFC) is positively correlated with the accuracy of behavior guided by working memory (Goldman-Rakic, 1992; Goldman-Rakic, 1995; Pratt and Mizumori, 2001) and that this activity is regulated by DA D₁ receptors (Williams and Goldman-Rakic, 1995). Pharmacological blockade of DA D₁ receptors in the PFC disrupts choice behavior guided by working memory, after both short (Sawaguchi et al., 1990; Sawaguchi and Goldman-Rakic, 1994) and long delays (Seamans et al., 1998). Furthermore, it has been proposed that working memory function is optimized when DA D₁ receptor occupancy is within a critical range of an inverted U-shaped function (Zahrt et al., 1997; Arnsten, 1998; Floresco and Phillips, 2001). Increased DA efflux in the primate PFC has been observed during repeated correct performance of a brief 10 sec delayed alternation task (Watanabe et al., 1997). These findings indicate a general role for DA in working memory; however, it remains to be determined whether phasic changes in mesocortical DA efflux in the medial PFC (mPFC) occur during different phases (i.e., acquisition, delay, and retrieval) of a delayed response task. Further evidence for dopaminergic modulation of memory processes mediated by the frontal lobes could be provided by in vivo measures of DA efflux in the PFC during different phases of delayed memory task.

The present study used in vivo microdialysis to monitor changes in DA release in the mPFC of rats performing a delayed response task on a radial-arm maze, using a delay period comparatively longer than previous studies (30 min). Unlike many other delayed-response tasks, this protocol does not depend on the component of working memory related to short-term (10-60 sec) retention of information, mediated by local PFC circuits (Goldman-Rakic, 1995). In this regard it is important to note that a broader definition of working memory, originally put forth by Baddeley (1986), refers to a collection of processes that include the temporary storage of information, as well as executive functions that mediate the manipulation and retrieval of trial-unique information to guide action after both short (seconds) and longer (minutes to hours) delays (Mizumori et al., 1987; Baddeley, 1992; Floresco and Phillips, 2001). In the present task (Fig. 1) information concerning the future location of food reward is obtained by a rat during a discrete acquisition phase, and memory of this trial-unique information is used to locate food accurately in a retention phase (Seamans et al., 1998). Behavioral studies using reversible disconnection lesions have confirmed that this type of delayed memory task is subserved by a neural circuit linking the ventral hippocampus to the mPFC (Floresco et al., 1997, 1999). Blockade of DA D₁ receptors in the mPFC during retrieval, but not acquisition, selectively disrupts memory-based search behavior (Seamans et al., 1998). Accordingly, we hypothesize that a significant increase in DA release in the mPFC will occur during accurate recall of the correct location of food sources in a complex spatial environment, independent of the presence or absence of food.

A portion of these data formed part of a poster presentation at the Tenth Annual Conference of the Rotman Research Institute, Toronto, Canada, March 2000.

Materials and Methods

Apparatus. Foraging tests were conducted on an eight-arm radial maze, consisting of an octagonal center platform (51 cm in diameter, arm-to-arm) connected to eight equally spaced arms (40 × 13 cm). A wooden post (69 cm in length) was positioned between two of the arms, and a lever mechanism attached to the top of the post adjusted the tension on the dialysis tubing when a rat moved through the maze. A liquid swivel (Instech 375s) located at the end of the lever was connected directly to the tubing, permitting rats to unimpeded movement.

Surgical procedures and behavioral training. Male Long-Evans rats (280-350 gm) were anesthetized with xylazine and ketamine hydrochloride (7 and 100 mg/kg, i.p., respectively) and stereotaxically implanted with bilateral stainless steel guide cannulas (19 gauge, 15 mm, nitric acid-passivated) 1 mm below dura, over the mPFC (+3.0 mm anteroposterior and ± 0.6 mm mediolateral from bregma). An additional guide cannula mounted on the skull (training post) with dental acrylic was used to tether the dialysis assembly during training sessions and thereby habituate the animals to this device. A stainless steel coil protected the dialysis tubing from the training post to the liquid swivel.

After a 7 d postoperative recovery period, rats were food deprived to 85% of their free-feeding weight. The delayed spatial win-shift (SWSh) task used in the delayed working memory tests is diagrammed in Figure 1. In all experiments described below, rats were trained in the same manner (i.e., 30 min delays, food available during both training and recall phases) regardless of the manipulations that were conducted on microdialysis test days. On the first two days of training, rats were habituated to the maze, and on subsequent daily delayed SWSh training trials, they were habituated to being tethered to the dialysis assembly. On the recall test, rats were permitted as many arm choices as necessary to retrieve the four pellets, and errors were scored as entries into unbaited arms, with criterion performance of one error or less during the test phase. The initial delay between training and recall test phases was 5 min. This delay was extended to 30 min after a rat achieved criterion performance for 2 consecutive days.

When criterion performance at a 30 min delay was maintained for 3 consecutive days (∼25-30 d of training), a microdialysis probe was inserted into the mPFC via the guide cannula, and a microdialysis experiment was conducted on the following day. In one group of rats (n = 6), microdialysis samples were collected during a standard delayed SWSh trial, with food available during both the training phase and recall test phase. As a control for the effect of food consumption on DA efflux, dialysis samples were taken from a second group of rats (n = 6) tested with food available during the training phase, but not during the recall phase of the delayed SWSh task. In a third group (n = 7), dialysis samples were collected during the standard training phase of the SWSh, and again during recall following an extended 1 hr delay between training and test procedure. In the final experimental group (n = 7), dialysis samples were collected during the training phase and again during a recall test after an extended 6 hr delay. During all delay periods in the microdialysis experiments, rats were confined within a wooden box with an open top (30 × 30 × 45 cm) in the center of the maze.

Microdialysis and HPLC. Microdialysis probes were concentric in design with silica inlet-outlet lines. The active surface consisted of a semipermeable membrane 2.0 mm in length (340 μm outer diameter; 65,000 Dalton molecular weight cutoff; Filtral 12; Hospal, Neurnberg, Germany). Probes were flushed continuously at 1 μl/min with a modified Ringer's solution (in mm: 10 NaHPO4, 1.2 CaCl2, 3.0 KCl, 1.0 MgCl₂, and 147.0 NaCl, pH 7.4) using a 2.5 ml gas-tight syringe (Hamilton, Reno, NV) and a syringe pump (model 22; Harvard Apparatus, South Natick, MA). Typical in vivo recoveries of DA conducted at room temperature were 17% of a standard DA solution.

Analysis of DA content in mPFC dialysates involved separation by reverse-phase high-pressure liquid chromatography (HPLC) and quantification by electrochemical detection (ED). The system consisted of a GBC (Victoria, Australia) LC1120 HPLC pump, a Scientific Systems, Inc. (State College, PA) pulse damper (316 stainless steel model), a Rheodyne manual injector (model 9125; 20 μl injection loop), a Princeton (Cranbury, NJ) SPHER C₁₈ column (2 × 10⁰ mm; 60A 3 μm particles), an Antec Leyden (Leyden, The Netherlands) Links system and an Antec Intro detector with a VT-03 electrochemical flowcell (V_applied = +0.7 V). The mobile phase, a 67 mm sodium acetate buffer containing 2.15 mm sodium octyl sulfate and 0.07 mm EDTA, pH 3.5 (12% v/v methanol), flowed through the system at 0.2 ml/min. Waters Maxima software was used to collect and analyze chromatographic data. A three-point calibration curve of external DA standards was used to convert the area under the curve of DA peaks into concentration values.

Microdialysis experiments. Microdialysis probes were inserted (-4.5 mm below dura) via guide cannulas into the mPFC, 14-16 hr before the start of each dialysis experiment. Implanted animals remained confined overnight in the center of the maze, with a daily ration of food and ad libitum water. Dialysis samples were collected at 5 min intervals and analyzed immediately with HPLC. Baseline samples were collected while the rat was confined in the center of the maze. After baseline samples had stabilized (<5% change over four consecutive samples), the box was removed, thereby initiating the training phase of the SWSh task in which four open arms on the maze contained food. During the standard 30 min and extended delay conditions, dialysis samples were obtained at 5 min intervals throughout the 1 hr delay, the recall test phase, and for 20 min after completion of the test phase. In the 6 hr delay condition, samples were collected for 40 min after the training phase, after which no samples were taken until 15 min before the recall test phase. During both the training and test phases for all four conditions, rats were permitted to explore the maze for the entire 5 min sampling period, even if they had retrieved all four food pellets before that period was over. However, only entries into unbaited arms that occurred before retrieving the final pellet were scored as errors.

Data analysis. Neurochemical data were transformed into percentage of change from baseline (i.e., 100% representing the average concentration of the three samples preceding the final baseline sample). Neurochemical data from each of the four treatment conditions were analyzed using a one-way repeated measures ANOVA followed by the Dunnett method of multiple comparisons, using the baseline sample immediately before the training phase as the control sample. Behavioral data from each experiment (errors, latency) for both the training and test phases were analyzed using separate one-way repeated measures ANOVAs with training day (day before test day and test day) as the within-subjects factor. In addition, statistical comparisons across the experimental conditions for both neurochemical and error data were conducted using one-way between-subjects ANOVAs in conjunction with Dunnett's method of multiple comparisons, using the 30 min delay with food condition as the control.

Results

Discussion

The present results show that DA efflux in the mPFC increased in a phasic manner during both the acquisition and retrieval phases of a delayed response task, but returned to baseline levels during the delay period itself. It is unlikely that the increase in mesocortical DA release observed during the training phase plays a role in the encoding of information to be used during the recall test because we have shown previously that inactivation of the mPFC before the training phase does not disrupt retrieval of information during the recall test 30 min later (Seamans et al., 1995). Rather, we propose that the subsequent increase in PFC DA efflux that occurs during the recall test is of primary importance in the accurate retrieval of information, a contention supported by the fact that blockade of D₁ receptors in the PFC before the test phase disrupts performance (Seamans et al., 1998).

To date, the only evidence linking increased in vivo DA efflux in the PFC to accurate recall during a delayed response task used a 15 min sampling period with a repeated trials design in which monkeys received sequential testing consisting of a 20 min period of sensory cue guided responding, a 20 min rest period, and a 35 min session of delayed alternation responding, with a 5 sec delay (Watanabe et al., 1997). The present findings extend these earlier results in several important ways. First, the data show that DA efflux is elevated in the mPFC of the rat after an extended delay of 30 min and in a situation in which spatial memory for the correct location of food must be recalled and incorporated into an accurate plan to search specific arms on the maze. Second, inaccurate recall of the correct locations of food pellets was negatively correlated with the magnitude of DA efflux. Specifically, after a relatively short (30 min) delay, animals were able to use information stored in memory to locate food efficiently, and their accurate search behavior was associated with a robust increase in mesocortical DA efflux. In contrast, when the delay between acquisition and retrieval of information was extended to a period of up to 6 hr, animals tended to forget where food was located on the maze and made more erroneous responses that were coincident with no significant change in the extracellular levels of DA in the mPFC.

The increase in DA efflux in the mPFC was not related to the consumption of a primary reward during the retrieval phase because the magnitude of DA increase was comparable whether or not animals had access to food. Moreover, when the same quantity of food was consumed after an extended delay of 6 hr (when animals had forgotten where food was located), no significant increase in DA efflux was observed in the mPFC. Our finding that the magnitude of DA efflux in the mPFC was not affected by the omission of reward during the recall phase has important implications for the hypothesis that the activity of DA neurons encodes information about conditioned reward-predicting stimuli, and “prediction errors” of primary reward (Schultz, 1998). DA neurons in primates initially display increased firing whenever rewards are presented unexpectedly, as is often the case during the initial phase of learning. Responding subsequently shifts to stimuli that predict reward and the activity of DA neurons is inhibited when reward is omitted (Ljungberg et al., 1992; Schultz et al., 1993; Schultz, 1998). The sustained elevation of DA during a 5 min period of reward omission observed in the present study stands in contrast to this inhibition of DA neuron activity (Hollerman and Schultz, 1998). Thus, whereas omission of an expected reward may be associated with a decrease in firing of individual DA neurons (Schultz, 1998), it is apparent that implementation of an accurate search strategy for food can override this effect and evoke increases in mesocortical DA release. Reconciliation of microdialysis data from the rat mPFC with electrophysiological data from the primate mesencephalon may come from the recent observation of a sustained increase in DA neuron activity related to uncertainty about motivationally relevant stimuli (Fiorillo et al., 2003). The uncertainty accompanying the search for food in a complex environment may serve as an optimal condition for evoking tonic firing of DA neurons, which in turn would increase the efflux of DA in terminal regions such as the mPFC.

Previous microdialysis studies of feeding behavior have shown that consumption of food or exposure to food odors during an anticipatory phase of feeding behavior is associated with a robust increase in PFC DA efflux (Bassareo and Di Chiara, 1997; Ahn and Phillips 1999, 2002). Given these observations, how can they be reconciled with the present finding that rats exposed to the 6 hr delay condition searched actively for food on the maze, consumed pellets once they were found, but showed no discernible increase in DA efflux in the PFC? Our recent studies of changes in DA efflux in the mPFC of the rat during sensory-specific satiety indicate that although sensory stimuli associated with a specific type of food can elicit a significant increase in DA efflux before and during its consumption, this relationship can be modified by recent experience such as feeding to satiety that can devalue this food (Ahn and Phillips, 2002). In a similar manner, contextual stimuli associated with the location of food may also have the capacity to activate the release of DA in the mPFC. Under circumstances in which specific associations between spatial cues and food were degraded by extended delays, as in the 6 hr condition in the present study, re-exposure to the maze may fail to elicit increased efflux of DA in the mPFC. This explanation emphasizes the probable relationship between memory-guided search behavior for food and mesocortical dopaminergic activity, rather than consumption of a primary reward.

As noted above, when an extended delay was inserted between the acquisition and retrieval phases of this delayed response task, rats made significantly more errors compared with their performance at a 30 min delay, and this increase in errors was negatively correlated with a decrease in the relative change of DA release in the mPFC. The question remains regarding whether the increase in retrieval errors during the extended delay conditions can be attributed to reduced levels of mesocortical DA or whether the reduction in DA efflux in the PFC is caused by the animals forgetting. Resolution of this question comes from neuropharmacological data from our laboratory (Floresco and Phillips, 2001). Using a similar, extended delay protocol, we found the disruption in working memory caused by an extended delay can be alleviated in part by infusion of a D₁ receptor agonist (SKF81297) into the mPFC before retrieval. The fact that pharmacological stimulation of D₁ receptors in the mPFC can restore working memory that is disrupted at a time when mesocortical DA release would be perturbed (i.e., by an extended delay) further supports the contention that the magnitude of DA release and the accuracy of working memory are causally linked. This finding, in combination with the pattern of results observed across the three delay conditions in the present study, indicate that the magnitude of DA efflux in the mPFC during the retrieval phase of a delayed response task is predictive of the accuracy of recall of baited arms, with lower levels of DA efflux associated with poorer performance.

Rather than construing mesocortical DA efflux as an index of the presence or absence of primary reward, we propose that it is a neurochemical correlate of cortical arousal that can facilitate the recall of information regarding the availability and location of a reward. In the context of working memory, a phasic increase in DA efflux in the mPFC during recall would ensure an appropriate level of DA D₁ receptor activation essential for the retrieval of trial-unique information by executive processes in the mPFC, which in turn guides behavioral selection during delayed response tasks (Sawaguchi et al., 1990; Sawaguchi and Goldman-Rakic, 1994; Williams and Goldman-Rakic, 1995; Seamans et al., 1998; Robbins, 2000). Although the time scales differ, the relationship between the magnitude of DA efflux and the accuracy of correct choices during recall tests parallels findings in electrophysiological studies of single-unit activity in the PFC during delayed memory tasks. In these experiments, robust increases in neural activity during either the delay period or responding are associated with accurate memory, whereas low levels of firing are accompanied by erroneous responses (Goldman-Rakic, 1992, 1995; Pratt and Mizumori, 2001).

Of relevance to the present study are anatomical data indicating that pyramidal neurons in the mPFC (including those which receive hippocampal inputs) send glutamatergic projections to DA neurons in the ventral tegmental area, which in turn are connected in a reciprocal manner to the PFC (Christie et al., 1985; Jay et al., 1995; Carr and Sesack, 2000; Floresco and Grace, 2003). These findings suggest that both the increase in task-related firing of PFC neurons and the enhanced efflux of mesocortical DA may be interrelated. Specifically, increased neural activity in the mPFC essential for efficient search behavior may in turn initiate corresponding changes in PFC DA efflux. Support for this conjecture is provided by data from our laboratory (Taepavarapruk and Phillips, 2001) and others (Gurden et al., 2000) confirming that stimulation of hippocampal afferents can increase DA efflux in the mPFC, an effect that is mediated by reciprocal connections between the PFC and ventral tegmental area (Taepavarapruk and Phillips, 2001). In light of these data, we propose that activity within a neural circuit linking the ventral hippocampus to the mPFC, which in turn engages the mesocorticolimbic DA projection, forms an essential substrate for working memory retrieval across short and long delays. The present data also imply that accuracy of memory-based search behavior is dependent on an appropriate level of DA in the mPFC, and that failure to recall the correct locations of food (as may occur when task demands increase) may be attributed in part to a perturbation in DA release that normally occurs during retrieval. Accordingly, drugs that serve as DA agonists, especially DA D₁ agonists (Floresco and Phillips, 2001), may provide pharmacotherapy for certain forms of dementia related to loss of function in the PFC.