Consolidation of learning strategies during spatial working memory task requires protein synthesis in the prefrontal cortex

Abstract

Working memory is a temporary memory store where information is held briefly until the appropriate behavior is produced. However, the improvement in the performance of working memory tasks with practice over days points to the existence of a long-lasting component associated with learning strategies that lead to optimal performance. Here we show that the improvement in the performance of mice in a radial maze working memory task required the integrity of the medial prefrontal cortex (mPFC). We further demonstrate that this improvement of working memory performance requires the synthesis of de novo proteins in the mPFC. We suggest that in addition to storing memory briefly the mPFC is also involved in the consolidation and storage of the long-term learning strategies used in working memory.

More than half a century ago, Hebb (1) proposed a dual-trace theory of memory in which he distinguished between short-term memory, a memory of recent events, and long-term memory, a memory of events that occurred in the past. Hebb proposed that the reverberating circuit of neuronal activity underlies short-term memory and that stabilization of this activity produces long-term memory. This dual-trace theory of memory found further support in the finding that inhibition of protein synthesis with antibiotics did not prevent the animals from learning tasks but disrupted their long-term memory (for example, see ref. 2). It is now widely accepted that there are at least two stages of memory: (i) short-term memory that temporarily stores information on the basis of changes in preexisting connections due to covalent modifications of preexisting proteins (3) and (ii) long-term memory that stores this information more permanently through the growth of new connections as a result of transcription and translation of certain genes, a process called consolidation (4, 5). A specialized form of short-term memory, called “working memory,” was introduced by the cognitive psychologists Baddeley and Hitch (6) to describe the memory that temporarily stores information provided by an environmental cue for the execution of an act in the near term. According to Baddeley and Hitch (6), working memory is the cognitive process that allows moment-to-moment perceptions across time to be integrated, rehearsed, and combined with simultaneous access to archival information about past experience, actions, or knowledge.

In primates and rodents, working memory is usually tested in a variety of moment-to-moment tasks in which sensory information is held in memory until a decision is made and the appropriate behavior is produced. However, a hallmark of working memory is that there is improvement in the performance of tasks over days with practice. There is a learning to learn with an improvement in strategy to perform the particular working memory task (7). In other words, while performing a working memory task, the subject is also learning a strategy that leads to optimal efficiency to perform on that task in the long-term. We have therefore set out to test the idea that learning strategies for working memory may have a long-term component that resembles long-term memory, and that their encoding may share the same properties of other well studied long-term memory processes, i.e., genes expression and protein synthesis.

There is now strong evidence that the prefrontal cortex plays a crucial role in working memory. Lesions of the prefrontal cortex produce a deficit in working memory in both primates and rodents (8–12). In addition, electrophysiological studies have found that neurons in the prefrontal cortex are active during the performance of working memory tasks (13–15) and also track the past and predict the future of performance (16). Beside its role in working memory, the prefrontal cortex is also involved in long-term memory processes (17–20), and patients with damage to the prefrontal cortex have impairments in the strategic utilization of memory (21).

In the present study, we explored the idea that the encoding of learning strategies needed for moment-to-moment working memory requires de novo proteins in the prefrontal cortex. We chose the mouse as animal model in our study because it offers advantages for a molecular–genetic approach to the study of long-term components of working memory. We found that the performance of mice in the radial maze, delayed nonmatching-to-place (DNMTP) working memory task with retroactive interference improved over days, and that this improvement required the prefrontal cortex. We next demonstrated that this improvement required the synthesis of new proteins in the prefrontal cortex.

Results

Effects of Medial Prefrontal Cortex (mPFC) Lesions on the DNMTP Task Without Retroactive Interference.

Mice were first trained to master the nonmatching-to-place (NMTP) rule in which the forced visit to one of the arms and its subsequent presentation in choice with an adjacent one are separated with a delay of 1–2 s. Then we assessed their working memory by varying the delay between the forced and choice runs (Fig. 1A). As depicted in Fig. 1B, both groups of mice learned the NMTP rule with the same rate and reached a plateau of 90% at the end of training. Analysis of the latency for each mouse to finish the session throughout the 7 days of training did not reveal any difference between the sham and PFC-lesioned mice (16.65 ± 4.66 and 19.95 ± 7.67 min, respectively) or a group × latency interaction indicating that the lesions did not produce any locomotor deficit. When a delay of 30 s was interposed between the forced run and the choice run, both groups showed a drop in their performance as revealed by an ANOVA on the last day of training and the 3-day mean with the delay of 30 s [F (1, 11) = 80.10, P < 0.001]. There was a difference between the two groups [F (1, 11) = 14.69, P < 0.01] and the group × session interaction was significant [F (1, 11) = 12.34, P < 0.01]. However, the performance of the sham group dropped to only 82%, which is still above the criterion of 75% of correct choice, whereas the performance of the lesioned group dropped to a level <75% (65%). After a rest of 1 week, the same mice were retrained with the NMTP rule because the data revealed this retraining was necessary. Once again, both groups mastered the rule equally well and reached a performance of 95% of correct choice. At the end of training, mice were given two daily sessions with delays of 15 and 30 s, respectively. A repeated-measures ANOVA performed on the last session of training, the session with 15-s delay, and the session with 30-s delay revealed that, overall, both groups showed a drop in performance with the increase in the delay [F (2, 22) = 87.18, P < 0.001]. There was no difference between the two groups, but the group × session interaction was significant [F (2, 22) = 12.20, P < 0.001]. A simple main effects analysis showed that the two groups significantly differed on the session with the 30-s delay (P < 0.01) but not on the session with the 15-s delay. Although the performance of the sham group was ≈79% at 30-s delay, the lesioned mice exhibited only a performance of 64% of correct choice. Collectively, these results showed clearly that the mPFC-lesioned mice were perfectly able to learn the NMTP rule in the eight-arm radial maze but presented a working memory deficit at a delay as short as 30 s.

Fig. 1.

Effects of mPFC lesions on the DNMTP task without retroactive interference. (A) Behavioral design. (B) Evolution of the percentage of correct response rates (mean ± SEM) during the DNMTP task without retroactive interference in sham- and mPFC-lesioned mice. Mice were first trained to master the NMTP rule with a very short delay (1–2 s), which was subsequently increased to 15 and 30 s. The break in the abscissa indicates a rest period of 1 week.

Effects of mPFC Lesions on the DNMTP Task with Retroactive Interference.

As depicted in Fig. 2C, the mPFC-lesioned mice performed poorly on the DNMTP working memory task (Fig. 2A) in which the delay between the forced run and the choice run was occupied by interposing one visit of another arm (retroactive interference) and failed to learn this task throughout the 10 daily sessions of training, whereas the sham-operated mice presented a progressive learning curve and reached the criterion of 81% of correct response by the sixth session of training. Repeated-measures ANOVA performed on the 10 sessions of training revealed a group effect [F (1, 14) = 51.2, P < 0.001] and a session effect [F (9, 126) = 5.0, P < 0.001], and the group × session term was significant [F (9, 126) = 2.0, P < 0.05]. Simple main effect tests revealed that only the sham group showed improved performance over the 10 sessions of training (P < 0.001), whereas the lesioned group continued to perform around the chance level.

Fig. 2.

Effects of mPFC lesions on the DNMTP task with retroactive interference and on the two-choice spatial discrimination task. (A and B) Behavioral designs. (C) Evolution of percent correct responses rates (mean ± SEM) during the DNMTP task with retroactive interference in sham- and mPFC-lesioned mice. (Dashed line indicates the chance level.) (D) Evolution of percent correct response rates (mean ± SEM) during the two-choice spatial discrimination task.

When next trained with the two-choice spatial discrimination task in the radial maze (Fig. 2B), both groups of mice showed similar curves of learning. Their performance improved over the seven daily sessions of training, and all animals reached the criterion of 80% of correct choice by the end of training (Fig. 2D).

Nissl staining of mice with lesions of the mPFC revealed extensive damage of the prelimbic and infralimbic subdivisions of the mPFC with occasional encroachment of the anterior cingulate cortex. A complete loss of neurons associated with a glial proliferation was observed in the area of the lesion (Fig. 3B). The anteroposterior extent of the lesions varied between 480 and 960 μm (mean 663 μm). In most cases, the lesions began between the frontal planes 2.71 and 2.22 of the atlas of Paxinos and Franklin (22). The posterior limit of the lesions was observed between the planes 1.70 and 1.50.

Fig. 3.

Representative photomicrographs of coronal sections showing a sham lesion (A) and an mPFC lesion (B). (Scale bars: 1 mm.)

Effects of mPFC Anisomycin (ANI) Injections on the DNMTP Task with Retroactive Interference.

We established the dose of 8 μg injected in each mPFC as the optimal dose to use with repeated injections without causing seizures. As a first step, we determined the degree of inhibition of newly synthesized proteins in the mPFC by the dose of 8 μg of ANI. Protein synthesis was quantified by the degree of incorporation of [³⁵S]methionine into proteins. The data are summarized in Fig. 4B. ANI injected 40 min before killing inhibited up to 83% of protein synthesis in the mPFC.

Fig. 4.

Effects of ANI injections on the DNMTP task with retroactive interference. (A) Photomicrograph showing cannula placements within the mPFC. (Scale bar: 1 mm.) (B) Mean percentage (±SEM) of incorporation of [³⁵S]methionine in the proteins extracted from the mPFC tissue of noninjected mice (NI), mice injected with saline (³⁵S/S), and mice injected with ANI (8 μg/0.2 μl per side) 20 min before the administration of the isotope (³⁵S/ANI). (C and E) Localization of cannula tips within the mPFC of mice injected with 8 and 64 μg/0.2 μl per side of ANI, respectively, and of mice injected with physiological saline (modified from ref. 22). (D) Evolution of the percentage of correct response rates (mean ± SEM) during the DNMTP task with retroactive interference in mice injected with saline and mice injected with ANI (8 μg/0.2 μl per side). Mice trained under ANI treatment were retrained 1 week later without injections (sessions 8–10). (F) Effects of ANI injections (64 μg/0.2 μl per side) in the mPFC of mice during the plateau phase of performance on the DNMTP task. The injections were performed 4 and 24 h after the end of the last session of training. Shown are mean ± SEM.

As illustrated in Fig. 4D, mice injected in the mPFC with 8 μg per side of ANI each day at the end of training performed poorly on the DNMTP working memory task and failed to reach a criterion of 81% of correct response at the last session of training. The vehicle-treated mice, however, presented a progressive learning curve and reached the criterion of 81% of correct response by the fourth session of training. Repeated-measures ANOVA performed on the seven sessions of training revealed a group effect [F (1, 15) = 49.0, P < 0.001] and a session effect [F (6, 90) = 15.9, P < 0.001], and the group × session term was significant [F (6, 90) = 8.4, P < 0.001]. Analysis of the latency for each mouse to finish the session throughout the 7 days of training did not reveal any difference between the saline- and ANI-treated mice (23.45 and 24.46 min, respectively) or a group × latency interaction, indicating that ANI administration at the end of each training session did not cause any motor deficit.

To check that the deficit observed in ANI-treated mice was due to the inability of the mice to consolidate memory rather than to tissue damage induced by the drug, the ANI-treated mice were given 1 week of recovery after the last session of training and then retrained with the same task without any injection. As depicted in Fig. 4D, the same mice that showed a deficit in mastering the DNMTP task under ANI treatment were able to show a normal learning curve when protein synthesis normalized and quickly reached the level of performance of the vehicle-treated mice. Statistical analysis of the performance of the ANI group obtained on days 2 and 3 of training without injections and the performance of the vehicle group on days 6 and 7 of training revealed no difference between the two groups [F (1, 15) = 0.11; not significant]. Thus, the deficit in performing on the DNMTP task observed in mice treated with ANI during training was due to their inability to consolidate memory rather than to tissue damage.

To verify whether postretrieval administration of ANI in the mPFC leads to amnesia, a group of naive mice was trained to master the DNMTP task during seven daily sessions, and at the end of the last session of training, they were injected bilaterally in the mPFC either with vehicle or 64 μg of ANI. As illustrated in Fig. 4F, ANI treatment did not affect the retrieval of acquired DNMTP when mice were tested 4 and 24 h after the injections [F (1, 12) = 0.02; not significant]. This result further indicates that ANI injections in the mPFC did not induce motor deficits.

Discussion

In this study, we have explored the idea that the encoding of learning strategies needed for moment-to-moment working memory is a form of long-term memory that requires de novo proteins in the prefrontal cortex. Our data showed clearly that the performance of mice in the DNMTP working memory task with retroactive interference in the radial maze improved over days and that this improvement required the integrity of the prefrontal cortex. Furthermore, and more important, we found that this improvement required the synthesis of new proteins in the prefrontal cortex as evidenced by the inhibition of this improvement by local injections of the protein synthesis inhibitor ANI.

Ibotenic acid lesions of the medial wall of the prefrontal cortex in these mice completely blocked the ability to perform the DNMTP working memory task. Unlike the sham-operated mice, the PFC-lesioned group did not show any improvement in their performance throughout the 10 days of training. In this eight-arm radial maze demanding task, the mice were required to recognize and choose a new arm (position) from another one previously presented before the delay, which was occupied by interposing one visit to a different arm that constituted a source of retroactive interference. The animals had to learn the NMTP rule through repeated daily sessions to be able to process specific spatial information and store it in their working memory. It is possible that the deficit observed in PFC-lesioned mice reflected a general learning impairment. In the two-choice spatial discrimination task, however, the lesioned mice did not show any learning deficit, and their performance was comparable to that of the sham group. Furthermore, mice with lesions of the mPFC were able to master the nonmatching-to-position rule with the same rate as their control group did. There was no evidence of a lesion-induced deficit in locomotor activity, and similar findings were reported in two recent lesion studies (23, 24). Taken together, these findings indicate that lesions of the mPFC did not compromise processing of attention or spatial information and did not produce a global learning, motivational, or motor deficit; rather, the lesions appear to affect specifically the ability of mice to learn the DNMTP task with proactive interference in the eight-arm radial maze. Because this task involves both working memory and complex spatial response and strategy selection mechanisms, the nature of the deficit that mice might show with lesions in the mPFC was not at all clear. But our data illustrate that mice with mPFC lesions had impaired working memory and confirm the importance of the mPFC for this working memory task in mice as well. This working memory impairment may have contributed to the inability of the lesioned mice to learn the DNMTP task with retroactive interference. Alternatively, the mPFC lesions may have prevented the mice from developing strategies to perform efficiently on the DNMTP task. This possibility is supported by the finding that the prefrontal cortex is critical for planning across long periods of time (16, 25, 26) and by the functional neuroimaging and neuropsychological studies that suggest a role of the PFC in using strategies to encode and retrieve memories and to cope with interference between similar events (27–29). Furthermore, the idea that the lesions may have prevented mice from learning the strategy necessary to perform the DNMTP task fits within the rule model proposed by Wise et al. (30) on the basis of their observation that deficits produced by prefrontal cortex lesions are attributed to the inability to use efficiently the appropriate strategy to perform correctly on a task. In any case, our data clearly show that learning performance in a more demanding eight-arm radial maze DNMTP task with retroactive interference improves over days with practice and that this improvement requires the integrity of the mPFC.

The number of errors in the DNMTP task with retroactive interference of sham-lesioned control mice decreased from chance to 15% in the 1st week of training, reflecting their ability to improve in performing on the task with practice. Thus with repeated training, mice learned how to use working memory to perform on the DNMTP task. In performing on the DNMTP task, mice were learning strategies that lead to optimal efficiency in the long-term. This finding is consistent with the idea that learning of strategies required to solve complex problems, such as the DNMTP task with retroactive interference, may have a long-lasting memory component that shares the cellular and molecular properties of other well studied long-term memory processes, i.e., gene expression and protein synthesis (4, 31, 32). In a recent paper, Huang et al. (33) reported that long-term potentiation (LTP) can occur in the rat mPFC by applying repeated tetanizations to slice preparations and that this LTP is blocked by ANI. Our data further demonstrate that consolidation of learning strategies in the DNMTP task with retroactive interference requires protein synthesis and that the mPFC is a critical site for this learning-induced de novo proteins. Once completely consolidated, this long-term memory becomes nonsensitive to protein synthesis inhibition. Collectively, these findings are consistent with neuropsychological and neuroimaging data involving the prefrontal cortex in long-term memory function (12, 19, 21, 34) and demonstrate that the mPFC is a site of consolidation and storage of learning strategies necessary for the efficient use of working memory. The findings further point to the existence of genes in the prefrontal cortex neurons that are important for some of the cognitive functions that depend on this area. Because a prefrontal dysfunction has been observed in schizophrenia, which is associated with working memory and long-term memory deficits (35), a detailed characterization of these genes and the molecular mechanisms of learning strategies, along with the pharmacological approach, could facilitate the assessment and treatment of individuals with this disorder.

Materials and Methods

Animals.

Male (3–5 months old) C57BL/6 mice from Taconic Farms (Germantown, NY) were used. The mice were individually housed in clear plastic cages in a vivarium maintained at 21°C and under a normal 12:12 light/dark cycle.

Surgery.

The mPFC (n = 18) was bilaterally destroyed by local injections of ibotenic acid. Under avertin anesthesia (2.5%, 0.5 ml per mouse, i.p.) and under stereotaxic control, a glass micropipette (tip diameter, 0.1 mm) was successively implanted in the mPFC at the following coordinates: 2.4 and 2.9 mm anterior to bregma, 0.4 mm lateral to the midline, and 2.3 mm below the skull surface. The micropipette was connected to a Hamilton 1-μl microsyringe that was secured firmly and attached to a stereotaxic carrier, and the controlled advancement of the syringe plunger was accomplished by using a 25-mm micrometer drive. This microinjector allowed delivering, in each side, 2 μg of ibotenic acid in 0.2 μl of the vehicle (sodium phosphate buffer, pH 7.2). By using the same procedure, 16 other mice were injected with the vehicle of the neurotoxin and served as sham-lesioned group.

For intracerebral microinjections, 44 mice were bilaterally implanted in the mPFC with a stainless steel guide cannula (Plastics One, Roanoke, VA) that was secured on the skull with stainless steel screws and dental cement, 1.0 mm above the center of the prelimbic subdivision (coordinates: 2.6 mm anterior to bregma, 0.4 mm lateral to the midline, and 1.3 mm below the skull surface). The guide cannulas were occluded with stainless steel pins to prevent clogging.

Mice were allowed a recovery period of 10 days before the behavioral tests and brain injections. All procedures involving animals were performed according to the guidelines of the Animal Ethical Committee of Columbia University.

Chemicals.

Ibotenic acid (Sigma, St. Louis, MO) was dissolved in sodium phosphate buffer, pH, 7.2, and brought to the final concentration of 10 μg/μl. ANI (Sigma) was dissolved in an equimolar concentration of HCl, adjusted to pH 7.2 with NaOH, and brought to the final concentrations of 8 and 64 μg/μl in physiological saline. [³⁵S]Methionine express protein labeling mix (Perkin–Elmer Life Science, Boston, MA) was used at the concentration of 3 μCi/0.2 μl per side in physiological saline.

Drug Administration.

At the time of intracranial microinjection, the stylus was removed, and a stainless steel cannula, connected to a Hamilton microsyringe via tubing, was inserted in the guide. The tip of the injection cannula protruded 1.0 mm beyond that of the guide. Injections of ANI (8 μg/0.2 μl per side and 64 μg/0.2 μl per side, dissolved in physiological saline) or vehicle were performed in freely moving mice at the rate of 0.1 μl/min at the end of each training session or at the end of the last training session.

Molecular Procedures.

Inhibition of protein synthesis was assessed by administration of [³⁵S]methionine (3 μCi/0.2 μl per side) in the mPFC 20 min before the animal was killed. ANI (8 μg/0.2 μl per side) was injected in each side 20 min before the administration of the isotope. The mPFC was dissected (approximately a block of 2.00 mm rostrocaudal, 1.00 mm lateromedial, and 2.00 mm dorsoventral) from mice with no injection (n = 5), and mice that received either injection of ANI (n = 5) or saline (n = 6), homogenized in lysis buffer (65 mM Tris·HCl/8 M urea/4% SDS, pH 6.8), and the proteins were precipitated by adding 4 vol of ice-cold isopropanol and incubated at −20°C for 4 h. Protein precipitates were isolated by centrifuging at 14,000 × g for 30 min, and residual acetone was removed by air-drying. The radioactivity in the total sample supernatant and pellet was determined by using a scintillation counter. Newly synthesized proteins were measured by quantifying the degree of incorporation of [³⁵S]methionine in the pellet of the homogenate. The percentage of inhibition of newly synthesized proteins was calculated by dividing the quantity of synthesized proteins after ANI injection by the quantity of synthesized proteins after saline injection.

Behavioral Procedures.

Apparatus.

The apparatus used for behavioral testing was an elevated eight-arm radial maze made of white Plexiglas and located in a quiet testing room enriched with distal spatial cues. The maze was adapted from Marighetto et al. (36) and consisted of an octagonal central platform from which radiated eight symmetrical arms. A door was mounted at the entrance to each arm from the central platform, and a food well was installed at the end of each arm. A manual device controlled door movements by using air pressure.

Habituation.

Mice were put on a food restriction diet so that their individual body weights were reduced to and maintained at 85% of the ad libitum weights. Before behavioral testing, they were habituated to the apparatus over 2 days by allowing them free exploration until all of the food pellets in the food well of each arm were collected.

DNMTP Working Memory Tasks.

In the first task, mice were trained to master the NMTP rule in which the forced visit to one of the arms and its subsequent presentation in choice with an adjacent arm are separated with a delay of 1–2 s. Each daily session was composed of a total of 16 trials. Once mice attained the minimum criterion of 83% of correct choice on 2 consecutive days, they were submitted to the DNMTP task in which the mnemonic demand was increased by increasing the delay between the forced run and the choice run (DNMTP without retroactive interference). Two delays of 15 and 30 s were used to evaluate the working memory performance during which the mouse was held on the central platform of the maze before the choice phase.

In the second task, the mice were trained to perform the DNMTP working memory with retroactive interference. This task assessed the animal's ability to recognize a new place sample (arm) from another previously made familiar by a single presentation. The delay between the forced run and the choice run was occupied by interposing one visit that constituted a source of retroactive interference, and normal mice mastered this task within 1 week of training (36–38). Each trial consisted in a study phase (forced run) followed by a test phase (choice run). During the study phase, mice were given two sequential forced runs to two different baited arms. After entering the last arm and returning to the central platform, the first visited arm and an adjacent one were opened simultaneously (first choice). Once the animal had chosen one of the two arms and had returned to the platform, the second arm visited in the study phase and an adjacent novel one were opened simultaneously (second choice). Once the animal made a choice, the doors of the other open arms were closed. The trial ended when the animal returned to the platform. In this test phase, a correct choice allowed the mouse to obtain a food pellet reward only in the arm that had not been previously visited during the study phase. Mice were given seven daily sessions of eight trials. In each trial, they were subjected to two successive choices. Thus in these conditions, recognition was required after one interposing visit (forced or choice run) between a forced run and its relevant choice run. Between trials, the mouse was confined in the central platform for 30 s.

To avoid any locomotor or position strategy, sequences of forced runs and choice samples were selected in a quasi-random manner with each problem counterbalanced by an opposite one. For example, for a given problem, if the correct arm is located to the left of the previously visited arm, the mouse will be next confronted with a problem for which the correct arm choice will be located on the adjacent right side of the initial sample. Consequently, if a given mouse systematically chose either the left or the right arm, its performance will be only at chance level (50%).

Concurrent Spatial Discrimination.

For the two-choice spatial discrimination task, each animal was separately assigned a different set of two pairs of adjacent arms made up each from one baited arm and one nonbaited arm. The minimum and maximum angles between the baited arms of the two pairs were 135° and 225°, respectively. In addition, it was ensured that if the positive arm of one pair was on the left (or right), then the positive arm of the other pair had to be on the right (or left). Mice were trained to learn this task over 7 consecutive daily sessions, each composed of 20 trials. Each trial started by opening the doors of the arms of a given pair while the mouse was on the central platform. When the animal reached the end of an arm, the other arm was immediately closed. The pairs were presented successively in a pseudorandom order. The trial was finished as soon as the subject returned to the central platform where it was confined for 15 s before the next discrimination trial began. Discrimination performance was assessed by the percentage of correct choices recorded on each 20-trial session.

Histological Analysis.

At the completion of the studies, mice with lesions or cannulas in the mPFC were anesthetized with avertin and perfused transcardially with physiological saline followed by a 10% (vol/vol) formalin solution. The brains were removed and soaked in a 10% (vol/vol) formalin solution containing 20% (wt/vol) sucrose for several days. Subsequently, the brains were coronally sectioned at 40 μm with a freezing microtome, and the sections were collected on albumin-coated glass slides and stained with thionin. The extent of the lesions and the position of the cannula tracks were examined under a light microscope and reconstructed on the appropriate frontal planes of the mouse brain atlas of Paxinos and Franklin (22).

Statistical Analysis.

Data were compared by using standard ANOVAs. Individual comparisons were evaluated with simple main effects tests or a Student's t test.

Abbreviations

ANI

anisomycin

DNMTP

delayed nonmatching-to-place

mPFC

medial prefrontal cortex

NMTP

nonmatching-to-place.