Medial Prefrontal Lesions Impair Performance in an Operant Delayed Non-Match to Sample Working Memory Task

Abstract

Cognitive functions, such as working memory, are disrupted in most psychiatric disorders. Many of these processes are believed to depend on the medial prefrontal cortex (mPFC). Traditionally, maze-based behavioral tasks, which have a strong exploratory component, have been used to study the role of the mPFC in working memory in mice. In maze tasks, mice navigate through the environment and require a significant amount of time to complete each trial, thereby limiting the number of trials that can be run per day. Here, we show that an operant-based delayed non-match to sample (DNMS) working memory task, with shorter trial lengths and a smaller exploratory component, is also mPFC-dependent. We created excitotoxic lesions in the mPFC of mice and found impairments in both the acquisition of the task, with no delay, and in the performance with delays introduced. Importantly, we saw no differences in trial length, reward collection, or lever-press latencies, indicating that the difference in performance was not due to a change in motivation or mobility. Using this operant DNMS task will facilitate the analysis of working memory and improve our understanding of the physiology and circuit mechanisms underlying this cognitive process.

Introduction

Cognitive deficits are a hallmark of most, if not all, psychiatric disorders. Cognitive processes, which range from attention and working memory to social cognition and the use of language, can be disrupted in psychotic, stress-related, developmental, and mood disorders (Kolb & Whishaw, 1983; Millan et al., 2012; Weinberger & Berman, 1996). Importantly, these symptoms are often predictors of long-term functional outcomes (Green, Kern, Braff, & Mintz, 2000). However, most currently available therapeutics target other symptoms of these disorders, leaving the cognitive impairments untouched, or even worse (Hill, Bishop, Palumbo, & Sweeney, 2010; Millan, 2006). Given this current therapeutic limitation, it is incumbent upon us to better understand the underlying neurobiology of these cognitive behaviors in order to develop more effective treatments.

Decades of research have identified the prefrontal cortex (PFC), an evolutionarily conserved part of the frontal lobe, as an important center for cognitive function (Frontal lobe function and dysfunction, 1991). Several reports demonstrated a striking resemblance between the cognitive deficits observed in patients with frontal lesions and those deficits seen in schizophrenia, suggesting alterations in the PFC may play a role in cognitive symptoms seen in this disorder (Kolb & Whishaw, 1983; Kraepelin, Robertson, & Barclay, 1919). Therefore, the study of the roles that prefrontal circuits play in cognitive functions is essential to furthering our understanding of the pathophysiology of schizophrenia, or any disorder with alterations in PFC circuitry.

In addition to patient-based research, much of our current knowledge of the role of PFC circuitry in behavior is based on animal studies. Using primates and rodents, researchers have manipulated specific brain circuits to determine their role in cognitive behaviors. To assess spatial working memory in rodents, a delayed non-match to sample (DNMS) T-maze task is frequently used (Bolkan et al., 2017; Kellendonk et al., 2006; Parnaudeau, Bolkan, & Kellendonk, 2018; Parnaudeau et al., 2013). In this task, the animal is trained to run from the start arm of a T-shaped enclosure to the available open arm during the initial sample phase, where it receives a reward. The animal then returns to the start arm, where it is held for a variable delay period. In the subsequent choice phase, both arms of the maze are opened, and the animal must go to the opposite arm (“non-match”) from the one that was presented during the sample phase in order to receive a reward. Longer delays increase task difficulty and are accompanied by a decrease in performance.

Previous work in rodents has shown that this DNMS T-maze task is dependent on the medial PFC (mPFC) both during the acquisition of the task, when there is a short delay, and upon introduction of longer delays, which increasingly tax working memory (Aultman & Moghaddam, 2001; Bolkan et al., 2017; Granon, Vidal, Thinus-Blanc, Changeux, & Poucet, 1994; Kellendonk et al., 2006). During the introduction of longer delays, the inputs to the mPFC that contribute to different phases of this task have been further explored. Input from the mediodorsal nucleus of the thalamus (MD), a higher order thalamic nucleus has been shown to be important during the delay (maintenance) phase, while input from the ventral hippocampus (vHPC) is critical for the sample (encoding) phase (Bolkan et al., 2017; Spellman et al., 2015). In addition, activity in mPFC-to-MD projections is necessary during the choice (selection) phase (Bolkan et al., 2017; Parnaudeau et al., 2018).

These studies have provided important insight into the circuits involved in different aspects of spatial working memory. However, in the T-maze only a limited number of trials can be completed on a given day due to the long trial lengths. For example, in our hands the initial sample phase lasts 17 seconds on average, with individual delay lengths lasting up to 120 seconds. This long trial length, combined with the distances needed to travel to obtain each reward, limits the number of trials that can be completed in a given day. As a result, it is harder to detect small effect sizes when two different experimental manipulations are compared. Moreover, a reduced number of trials per day also limits the number of conditions or manipulations that can be introduced on a given day (e.g. testing multiple delay lengths in the same day). Finally, the ability to have more trials per day facilitates the analysis of in vivo physiological data collected during task performance. In addition to long trial lengths, the T-maze also has a strong explorative spatial component, with the animal navigating through long maze arms while encoding and subsequently retrieving a memory. With shorter time scales in a smaller, enclosed operant box, the working memory tested is less dependent on interference from navigating through the arms.

As an alternative to this traditional DNMS T-maze task, several operant versions have been developed, which are inherently less exploratory, allow many more trials per day to be conducted, and facilitate the simultaneous collection of data from a large number of animals. One version, developed by Rossi et al allows the mice to select one of two levers during the initial sample phase (Rossi et al., 2012). The levers are then removed, and subsequently reinserted after a given delay; the mouse must choose the lever it did not pick in the sample phase in order to correctly earn a reward in the choice phase. While performance in this task deteriorated in a delay-dependent manner and was impaired by mPFC lesion, there are several strategies the animal can develop to solve the task that avoid using working memory. For example, after selecting the initial sample lever, the mouse can immediately wait in front of the opposite lever until it appears, effectively overriding the need to utilize working memory during the delay time. To circumvent this limitation, an alternative task has been developed in which mice press an initial sample lever to initiate a delay phase, but then must make an entry to a noseport on the opposite wall at the end of the delay phase in order to trigger the presentation of both levers during the choice phase. However, while versions of this task have been shown to be dependent on the dorsomedial striatum (Akhlaghpour et al., 2016) and hippocampus (Goto & Ito, 2017), it is imperative to know whether it is also dependent on the mPFC.

In this study, we implemented an operant-based DNMS task, similar to that used by Akhlaghpour et al (Akhlaghpour et al., 2016) and Goto et al (Goto & Ito, 2017), in which mice performed as many as 160 trials per day, to establish its dependence on the mPFC. We found that mPFC lesions impaired both the acquisition of the task, which was done in the absence of a delay, and task performance after introduction of different delays. We further showed that trial length, reward collection, and lever-press latencies were unchanged by the lesion, indicating that an underlying decrease in motivation was not responsible for the impaired task performance in mPFC lesioned animals. Thus, the DNMS operant-based task will serve as a useful complement to the DNMS T-maze task, facilitating the study of prefrontal circuitry and physiology in working memory.

Method

Animals.

All experiments were carried out on male C57/Bl6 mice purchased from Jackson Laboratory (Stock #000664). Mice were aged 8 weeks at the start of experiments and housed under a 12-h light-dark cycle in a temperature-controlled environment with food and water available ad libitum. Mice were group housed with littermates (5 mice/cage). During behavioral training and testing, mice were food-restricted and maintained at 85% of their initial weight. All procedures were done in accordance with guidelines derived from and approved by the Institutional Animal Care and Use Committees at Columbia University and the New York State Psychiatric Institute.

Surgical procedures.

Mice were anesthetized with ketamine (10mg/ml) and xylazine (1mg/ml) and head-fixed in a stereotactic apparatus (Kopf). Mice were injected bilaterally into the mPFC with either ibotenic acid (Sigma-Aldrich, I2765), dissolved in ddH₂O at 10 mg/ml, or phosphate buffered saline (PBS), at a volume of 0.25 μl (0.1 μl/min). The ibotenic acid was stored at −20°C, and just prior to the injection was re-dissolved at 37°C. The mPFC coordinates used were: +1.8 AP, ±0.35 ML, −2.5 DV (skull at bregma).

Behavioral apparatus.

Eight identical operant-conditioning chambers (ENV-307A; Med Associates, Georgia, VT) were used. The chamber measured 15.24 cm long x 13.34 cm wide x 12.7 cm high. Each chamber was housed in a sound-attenuated box and equipped with two retractable levers (ENV-312-3M) on the front wall (the 13.34 cm side), with one milk dipper between them (ENV-302RM-S, Fig. 2a). The back wall contained one noseport (ENV-313M) directly opposite to the milk dipper. A 1.0-A house light was positioned directly above the noseport. A computer (COM-106-NV, Intel i5-7400) controlled and recorded all experimental events and responses via an interface (MED-SYST-16e-V). Med-PC V programs were used to administer and record all behavioral tasks.

Figure 2. Acquisition of an operant DNMS task is impaired by an mPFC lesion.

a) Layout of the operant box. Left: Front wall, containing a milk dipper and two levers, one on either side of the dipper. Center: Back wall, containing a noseport and a 1.0-amp house light. Right: Top view of the operant box. The milk dipper and levers on the front wall are represented on the right side of the image, and the noseport on the back wall is represented on the left side of the image. b) Schematic illustration of the trial sequence for the acquisition of the task, including a 0-second delay. c) Performance of sham (light circles) and lesion (dark squares) groups over the 19 days of acquisition indicated as the percentage of correct trials on each day (n=16 sham mice, 17 lesion mice; two-way repeated-measures ANOVA (rmANOVA), main effect of lesion, F(1,31)=5.687, *p = 0.0234; time x lesion interaction, F(18,558)=1.951, *p = 0.0108). d) Number of days to reach the criterion of 3 consecutive days with a performance above 80% correct for sham (light) and lesion (dark) groups (two-tailed unpaired t-test, sham vs. lesion, t=3.048, df=31, **p = 0.0047). e) Mean length of each trial throughout acquisition (two-tailed unpaired t-test, t=0.1582, df=31, p=0.8753). f) Length of each trial for each day of acquisition. Dashed line represents the introduction on Day 6 of the imposed time limit for the second noseport entry. In the first five days of acquisition, there was unlimited time for the second noseport entry (two-way rmANOVA, main effect of time, F(4,110)=0.9300, p=0.4494; main effect of lesion, F(1,31)=0.2091, p=0.6507; time x lesion interaction, F(4,110)=0.3293, p=0.8578). Starting with Day 6, there was a 5-second time limit imposed on the second noseport entry (two-way rmANOVA, main effect of time, F(13,403)=1.240, p=0.2479; main effect of lesion, F(1,31)=0.004387, p=0.9476; time x lesion interaction, F(13,403)=1.064, p=0.3888). g) Latency between sample lever press (S) and choice lever press (C) throughout acquisition (two-tailed unpaired t-test, t=0.7222, df=31, p=0.4756). h) Latency between choice lever press and reward retrieval throughout acquisition (two-tailed unpaired t-test, t=0.7522, df=31, p=0.4576). i) Percentage of rewards awarded that were retrieved (two-tailed unpaired t-test, t=0.6377, df=31, p=0.5283). j) Percentage of trials that were completed, not aborted (two-tailed unpaired t-test, t=0.1954, df=31, p=0.8464).

Behavioral procedures.

Two weeks following the ibotenic acid injection, mice were gradually food restricted to 85% of their body weight. Mice were then shaped to the different parts of the operant task. First, the mice were given 2 days of dipper training, during which the mice were presented with the dipper containing 1 drop of evaporated milk (0.01 ml). Each day, the animals were given the opportunity to obtain 20 rewards in a maximum of 30 min with a random inter-trial interval (ITI), averaging 5 seconds. For the next 3 days, the animals were trained to associate a lever press with a milk reward. Each day, the mice were given a maximum of 60 minutes with each retractable lever, baited with High-Calorie Nutritional Gel (Tomlyn), to receive a maximum of 60 rewarded lever presses on each side. Every second trial, a 10-second ITI was introduced. Next, the mice were given one day during which each lever was presented 30 times in a pseudo-random order to receive a maximum of 60 rewards. For this experiment, pseudo-random refers to a random distribution with the restriction that the same lever cannot be presented for more than 2 consecutive trials. In the final step of shaping, the noseport was introduced. Each trial began with an illuminated noseport. When the noseport was entered, one of the two levers would extend in a pseudo-random order, and a lever press would result in a milk reward, followed by a 5-second ITI. Each day, the animal could perform a maximum of 60 rewarded trials within a maximum of 60 minutes. After 4 days of noseport training, the animals began the acquisition stage of the behavior.

Acquisition was repeated on 19 consecutive days. Throughout the 19 days, the animals were given unlimited time to complete the required trials. Each trial began with the house light being turned on and an illuminated noseport to signal an initial noseport entry. The first noseport entry triggered the start of the sample lever presentation. During the sample phase, only one lever was presented in a pseudo-random order. After the sample lever press, the noseport was immediately re-illuminated (following a 0-second delay) signaling a second noseport entry. Following the second noseport entry, the choice phase began, and both levers were presented. If the animal pressed the opposite lever to the sample lever of that trial (non-match), the trial was recorded as “correct” and a dipper reward was given. If the animal pressed the same lever as the sample, the trial was recorded as “incorrect” and the dipper was not presented. This final step was followed by a 10-second ITI during which the house light was turned off.

During the first 5 days of acquisition, there were 120 trials per day total; 60 trials with each lever presented as the sample. Furthermore, the animal had unlimited time following the sample lever press for the second noseport entry.

For the subsequent 10 days of acquisition (120 trials per day), mice had a 5-second time limit in which to make the second noseport entry. This restriction allowed us to shape the animals’ behavior to ensure a standardized length of delay between subjects. If the animal did not make a noseport entry in the time allotted, the trial was aborted and was omitted from the calculations.

Finally, during the last 4 days of acquisition, the number of trials was increased to 160 trials. This allowed the animals to adjust to the longer days before the delays were introduced.

During the acquisition stage, all mice achieved a criterion level of performance, defined as 3 consecutive days above 80% correct.

Following acquisition, the delay stage began. In this stage, each trial had the same structure as during the acquisition stage, with one difference: a delay of 2, 4, 8 or 16 seconds was introduced between the sample lever press and the second noseport illumination. Each day every mouse was presented with a total of 160 trials with 40 trials of each delay condition randomly interspersed. This testing was repeated for 5 days. On the first day of testing, all animals performed poorly even at the shortest (2-second) delay (despite a previous high performance with 0-second delays), demonstrating the need for a short adjustment period to the new task parameters. The data from this adjustment period was therefore excluded, and the last 4 days, when the performance was more consistent, were taken together for analysis.

All behavior was conducted during the light cycle.

Statistics.

A two-way repeated measures ANOVA was used to assess significant overall effect of lesion and interactions between lesion and time during the acquisition stage or between lesion and delay length during the delay stage. Two-tailed t-tests were performed to compare the number of days to criterion for the lesioned and sham-lesioned groups, as well as other task characteristics.

Histology.

At the end of experimentation, mice were transcardially perfused with PBS followed by 4% PFA. Fixed tissue was then sectioned (40 μm coronal) using a vibratome and mounted on charged slides. The tissue was stained with Cresyl Violet (Sigma-Aldrich C5042) to target Nissl bodies. Eight slices for each animal spanning and extending past the mPFC (from AP +3.2 to −0.5 relative to bregma) were then examined with a brightfield light microscope (Zeiss) to assess the location and extent of the lesion, which was determined based on loss of cell density, accumulation of clumped Nissl staining from dead tissue and contraction of the gray matter(Hunt & Aggleton, 1998) (Fig. 1c).

Figure 1. Experimental Timeline and Extent of Lesion.

a) Timeline of experimental procedures. b) Schematic representation of the maximal (light) and minimal (dark) extent of damage caused by the mPFC ibotenic acid injection in coronal slices. c) Example Nissl staining from a sham (right) and lesion (left) coronal slice. Dashed lines outline area with lower cell density and accumulation of Nissl found with dead tissue. Arrowheads indicate the shift of the white matter tract as a result of contraction of the medial regions. d) Higher magnification of example lesion coronal slice from c. Solid outlines show control region borders from the Paxinos and Watson Mouse Brain Atlas. Dashed lines outline the shifted border of the white matter tract.

Data availability.

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Code availability.

Med-PC V and Matlab code used for administering the behavior and analysis of the data that support the findings of this study is available from the corresponding author upon reasonable request.

Results

The goal of this study was to investigate the role of the mPFC in an operant-based DNMS working memory task. To address this question, we created an excitotoxic lesion of the structure using an injection of ibotenic acid, or conducted a sham surgery using an injection of phosphate buffered saline (PBS), before training the mice in the DNMS working memory task (Fig. 1a).

Confirmation of the lesion location.

Post-hoc histology showed that the lesioned region typically encompassed both the prelimbic and infralimbic portions of the mPFC. In some animals, the lesion also spread to anterior portions of the cingulate cortex (Cg1 and Cg2). No spread was seen to any of the motor or sensory cortices. The lesioned regions showed a number of distinct characteristics including 1) a loss of cell density, 2) an accumulation of non-cellular Nissl clumps, and 3) a contraction of the lesioned region, which were used to define the lesion boundaries (Fig. 1b, c, d).

Acquisition performance is impaired by the mPFC lesion.

Following a post-surgical recovery period, the mice began the first stages of training, which involved shaping in operant boxes with a reward milk dipper, levers, and noseport (Fig. 2a), as described in the Methods section. The shaping was learned similarly across the two groups. To learn the lever press, the sham group took 1.438 ± 1.031 days, and the lesion group took 1.412 ± 0.795 days. An unpaired t-test showed that the groups were not significantly different from one another (p=0.9363). Similarly, to learn to poke the noseport, the sham group took 1.688 ± 0.793 days, and the lesion group took 1.706 ± 0.686 days, with an unpaired t-test showing no significant difference between the groups (p=0.9436). The mice then learned the DNMS task with a 0-second delay; this period represents the acquisition stage of the task (Fig. 2b). Mice with an mPFC lesion were significantly slower to acquire the task, requiring more days of training to reach a criterion level of performance (Fig. 2c, d). While all animals began with a chance level of 50% performance, sham mice were able to learn the task more quickly. A two-way repeated measures ANOVA showed a significant main effect of lesion over time (p=0.0234) and a significant time x lesion interaction (p=0.0108). Bonferroni post-hoc analysis found the sham group performance to be significantly higher than the lesion group performance during training days 7 and 8 (day 7: p=0.0067; day 8: p=0.0191). While the sham mice were able to reach a criterion of 3 consecutive days above 80% performance in 8.56±2.58 days (mean ± standard deviation (SD)), the lesion mice took 12.24±4.12 days to reach the same level of performance.

Given this deficit in performance, we wanted to know whether other aspects of the behavior were affected. We found no difference between the groups for any other parameters measured. Total time to complete each trial did not change with training and was not different between groups (Fig, 2e, f). Similarly, the latency between pressing the sample and choice levers were also the same across both groups (Fig. 2g). This indicated that there was no gross motor impairment in the lesioned animals and suggested that the performance deficit of the lesioned group was not the result of a longer ‘experienced delay time’ in the task. There was also no change in the latency between the choice lever selection and reward retrieval (Fig. 2h), nor was there a difference in the percentage of rewards retrieved, with both groups retrieving over 99% of offered rewards (Fig. 2i). These similarities indicate that the difference in behavior is likely not a result in a change in motivation to earn the milk reward. Finally, the percentage of aborted trials (recorded when the animal failed to make the second noseport entry in the 5-second time limit) was below 5% for both groups, with no animal aborting more than 12% of trials (Fig. 2j), indicating that the animals in both groups had similar numbers of trials on each day in which to learn the task. Together, these data indicate that both the sham and lesion groups experienced the task similarly during the acquisition stage. However, the lesion group still took longer to acquire the task.

Delay performance is impaired by the mPFC lesion.

When every animal had reached the acquisition criterion, the delay testing began. Delays (2, 4, 8, or 16 seconds) were introduced between the sample and choice lever presentations (Fig. 3a). Each delay condition was randomly interspersed within a given day, and the testing was repeated over several days. The performance was summed across the last four days of testing for the data presented in Fig. 3. The baseline level of performance with a 0-second delay, taken as the average performance across the last 3 days of acquisition, was not significantly different between the groups (mean ± SD, sham: 90.65±5.93%, lesion: 89.90±5.74%). However, the performance with delays was impaired for the lesion group compared with the sham group (Fig. 3b). A two-way repeated measures ANOVA showed a significant main effect of the lesion across delays (p=0.0075) and a significant delay x lesion interaction (p=0.0014). This impairment was driven by the differences in performance in the 2- and 4-second delay conditions, where sham mice performed above or close to criterion level (2s: 83.27±7.84%; 4s: 77.41±6.31%), but lesion animals performed worse (2s: 74.84±9.38%; 4s: 65.01±11.35%). Bonferroni post-hoc analysis for each delay condition showed a significant difference between the groups at those two delays (p-values: 2s: 0.0209, 4s: 0.0002), and a trend-level difference between the groups’ performances at the 8-second delay condition (p=0.0919, sham: 69.25±10.36%, lesion: 62.34±9.91%), but no effect at the 16-second delay condition (p>0.9999, sham: 59.85±7.41%, lesion: 56.64±6.58%) where the performance for both groups was close to chance (50%).

Figure 3. Delay performance of an operant DNMS task is impaired by an mPFC lesion.

a) Schematic illustration of the trial sequence for the task, including randomly interleaved 2-, 4-, 8-, and 16-second delays. b) Performance of sham (light circles) and lesion (dark squares) groups indicated as the percentage of correct trials for each of the four delay lengths as well as baseline (‘Base’: mean performance from the last 3 days of acquisition; n=16 sham mice, 17 lesion mice; two-way rmANOVA, main effect of lesion, F(1,31)=8.179, **p = 0.0075; delay x lesion interaction, F(4,124)=4.729, **p = 0.0014; Bonferroni post-hoc corrected p-values: 2-s t=2.907, df=155, #p=0.0209; 4-s t=4.279, df=155, ###p=0.0002). c) Mean length of each trial for each delay length for sham (light) and lesion (dark) groups (two-way rmANOVA, main effect of lesion, F(1,31)=0.1205, p=0.7308; delay x lesion interaction, F(3,93)=0.05374, p=0.9835). d) Latency between sample lever press (S) and choice lever press (C) for each delay length (two-way rmANOVA, main effect of lesion, F(1,31)=0.5730, p=0.4548; delay x lesion interaction, F(3,93)=1.485, p=0.2238). e) Latency between choice lever press and reward retrieval throughout delays (two-tailed unpaired t-test, t=0.3436, df=32 p=0.7334). f) Percentage of rewards awarded that were retrieved (two-tailed unpaired t-test, t=0.6530, df=31, p=0.5186). g) Percentage of trials that were completed, not aborted (two-tailed unpaired t-test, t=0.6034, df=31, p=0.5507).

As with the acquisition stage, we also measured other parameters to assess whether the difference in performance might be accounted for by differences in behavior other than working memory. As before, trial length and the latency between sample and choice lever presses for each delay condition were the same between groups (Fig. 3c, d). Similarly, the latency to collect the reward and the percentage of rewards retrieved were similar between groups (Fig. 3e, f), as were the percentage of completed trials (Fig. 3g). Cumulatively, these findings suggest that mPFC lesioned mice show an impairment in performance of a DNMS working memory task, that is not due to motor or motivational impairments.

With the complicated structure of the task during this stage (i.e., having all four delay lengths randomly interleaved within one day), there was an improvement in performance for both the sham and lesion groups over time. Therefore, the difference in total performance observed between the two groups might have multiple explanations. It could be due to: 1) a difference in learning to cope with the newly introduced, variable delays, or 2) a difference in memory maintenance. To distinguish between these two possibilities, we evaluated the learning from day to day and within a day by analyzing performance in 10-trial blocks for each delay (Fig. 4). At all delay lengths, a two-way repeated measures ANOVA showed a significant main effect of time (p<0.0001 for all delays), indicating learning. In addition, there was a significant main effect of lesion at 2s and 4s, as was seen with the pooled data analysis (p-values: 2s: 0.0073, 4s: 0.0008) and a trend-level effect at 8s (p=0.0605), but no effect at 16s (p=0.2214). Crucially, there was no time x lesion interaction at the delays that revealed a difference between the groups (p-values: 2s: 0.2162, 4s: 9692, 8s: 9777). This analysis indicates that while there is an impairment in the performance of the lesion animals during the delays, this difference cannot be attributed to a difference in learning during the delay condition, but rather is explained by a difference in memory.

Figure 4. Delay performance is impaired by an mPFC lesion, but learning, over 10-trial blocks, is unaffected.

Performance for sham (light circles) and lesion (dark squares) groups over blocks of 10 trials for each delay length: a) 2-second delay (two-way rmANOVA, main effect of time, F(15,465)=3.328, ****p<0.0001; main effect of lesion, F(1,31)=8.262, **p = 0.0073; time x lesion interaction, F(15,465)=1.271, p = 0.2162); b) 4-second delay (two-way rmANOVA, main effect of time, F(15,465)=5.788, ****p<0.0001; main effect of lesion, F(1,31)=13.77, ***p = 0.0008; time x lesion interaction, F(15,465)=0.4330, p = 0.9692); c) 8-second delay (two-way rmANOVA, main effect of time, F(15,465)=5.183, ****p<0.0001; main effect of lesion, F(1,31)=3.75, p = 0.0605; time x lesion interaction, F(15,465)=0.4050, p = 0.9777); d) 16-second delay (two-way rmANOVA, main effect of time, F(15,465)=3.509, ****p<0.0001; main effect of lesion, F(1,31)=1.570, p = 0.2195; time x lesion interaction, F(15,465)=1.982, *p = 0.0151).

Both the acquisition and delay stages of this task demonstrated an impairment due to the mPFC lesion; however, the acquisition stage clearly represents a difference in learning whereas the delay stage appears to denote a difference in memory or task execution. While we were interested to discover how the performance in these two stages was differentially affected, we also wanted to know how they might be related. To that end, we analyzed each animal’s performance during both stages using a linear regression analysis of the total performance at each delay versus the acquisition, as measured by the number of days to reach the criterion (Fig. 5). This analysis revealed a significant though weak correlation between days to criterion and performance for each of the delays (2s: R²=0.2850, p=0.0014; 4s: R²=0.3783, p=0.0001; 8s: R²=0.2201, p=0.0059; 16s: R²=0.2943, p=0.0011). The correlation between days to criterion and delay performance demonstrates that the two stages are related. The R² values indicate that 22-37% of the variance in the delay performance can be explained by the days to criterion. Thus, the deficit in learning does not fully account for the difference in delay performance.

Figure 5. Correlations between acquisition and performance at each delay show a linear relationship.

Each animal’s delay performance, at each of the 4 delays, versus that animal’s acquisition performance, measured in days to criterion. At each delay, a linear regression was evaluated across all animals (sham light circles, lesion dark squares). All 4 delays demonstrated significant, weak linear relationships with the days to criterion (solid line with 95% confidence intervals in dashed lines). a) At 2-second delay, R²=0.2850, **p=0.0014; b) at 4-second delay, R²=0.3787, ****p=0.0001; c) at 8-second delay, R²=0.2201, **p=0.0059; d) at 16-second delay, R²=0.2943, **p=0.0011.

Discussion

The mPFC is required for acquisition and delay performance of an operant-based DNMS working memory task.

In rodents, working memory describes the ability to hold information online for short periods of time and then use it to accomplish a goal. Working memory can be assessed in a variety of different tasks where a delay separates the acquisition of information from the period in which the information must be used for the correct performance of the task. In this study, we used mice to investigate the role of the mPFC in an operant-based DNMS working memory task similar to one previously shown to be dependent on the striatum (Akhlaghpour et al., 2016) and the hippocampus (Goto & Ito, 2017). Although the mPFC, a structure that has been implicated in many cognitive functions and psychiatric disorders, has previously been shown to be required for other working memory tasks, we wished to validate this structure’s importance for the current operant-based task, which offers several advantages including a smaller exploratory component and a higher daily throughput. To that end, we explored the impact of an mPFC lesion on both learning and delay performance in the current operant working memory task in mice.

Our data demonstrate that lesioning the mPFC impaired both acquisition and delay performance in this operant task. During the acquisition of the task, where the trials contained a short (0s) delay between the sample and choice phases, all animals began at a chance level of performance (50% correct). As they continued to train in the task, the lesioned animals were slower to perform the task correctly than the sham animals. All mice eventually learned the task, reaching the criterion of 3 consecutive days above 80% correct performance; however, the lesioned animals took more days to reach this criterion. After all animals reached a similar high level of performance, delays were introduced. As expected, both sham and lesioned animals showed delay-dependent effects on performance, with accuracy dropping as the delay time increased. Additionally, lesioned animals showed impaired performance with the delays relative to sham animals that was most evident for the shorter delays of 2 and 4 seconds, when the sham animals continued to perform at a relatively high level. For the longer delays of 8 and 16 seconds, the sham animals performed very close to chance, likely making it difficult to see any additional deficit in the performance of the lesioned mice.

Other parameters of the task, such as the time taken to complete each trial or the percentage of total rewards collected, which might have reflected a change in mobility or motivation, were unaffected. This result indicates that the deficit in acquisition and delay task performance in the mPFC lesioned mice is not due to them experiencing longer effective delays or being less motivated to earn rewards. Moreover, as learning rates during the delay condition are comparable in lesioned versus non-lesioned mice a deficit in learning cannot explain the deficit observed under the delay condition. This interpretation is supported by the relatively weak correlation in the performances of individual animals during both stages. Thus, the deficit observed in the delay stage can at least partly be explained by a deficit in memory.

There are also costs associated with the DNMS operant task.

This operant DNMS task offers several advantages highlighted throughout this paper, such as shorter trial lengths and a smaller exploratory component. However, there is an associated cost of prolonged training. In the T-maze DNMS task, animals acquire the task within a week whereas in the operant task DNMS task, it took the animals three weeks. In addition, maze tasks heavily depend on locomotor exploration which is a more ethologically relevant behavior for a mouse than lever pressing. Performing the T-maze task, which requires spatial exploration of goal arms likely engages different, but also overlapping, neuronal circuitry than performing the operant task. In this context, it will be interesting to determine whether sample encoding is also dependent on the ventral hippocampal input to the mPFC in the operant DNMS task as it has been observed for the T-maze task (Abbas et al., 2018; Bolkan et al., 2017; Spellman et al., 2015).

Similar findings have been documented in other working memory tasks.

Previous work has demonstrated the importance of the mPFC in maze-based spatial working memory tasks in rodents. These studies showed that a lesion of the mPFC in either mice or rats impairs the acquisition of a DNMS T-maze task (Granon et al., 1994; Kellendonk et al., 2006), and that acute optogenetic silencing of projections to and from the mouse mPFC impairs performance of the task after delays are introduced (Bolkan et al., 2017). Similarly, delay performance in this task is impaired by an mPFC lesion in rats (Aultman & Moghaddam, 2001). Furthermore, studies disrupting the rat mPFC showed that this region is also important in several other more complicated maze-based working memory tasks, including a figure-eight maze (Yoon, Okada, Jung, & Kim, 2008) and a radial arm maze (Seamans, Floresco, & Phillips, 1995). These results, combined with those of the current study, demonstrate that the mPFC is an essential structure for the acquisition and performance of various working memory tasks with a non-match-to-sample structure.

Several studies also explored the role of the mPFC in maze-based working memory tasks that instead require delayed alternation. In the delayed alternation T-maze task, the animal starts by selecting one of two arms to visit. In all subsequent trials, the animal must choose the opposite arm from the one previously visited in order to receive a reward. While this task does require working memory, the alternation from trial to trial allows for an alternating strategy that would not work in the DNMS task, where for each trial the sample arm is newly chosen. Despite this difference, Larsen and Divac showed that performance in a delayed alternation T-maze task was impaired following a prefrontal lesion in rats, similar to what has been seen in DNMS T-maze tasks (Larsen & Divac, 1978). Therefore, it appears the mPFC is essential for both DNMS as well as delayed alternation based T-maze working memory tasks.

Other investigators have studied the role of the mPFC in an operant version of a delayed alternation task, finding that an mPFC lesion impairs performance (Dunnett, Nathwani, & Brasted, 1999; Rossi et al., 2012). For instance, in Rossi et al, an mPFC lesion in mice impaired both acquisition and delay performance, similar to our findings (Rossi et al., 2012). Of note, their operant task differed from the delayed alternation T-maze task of Larsen & Divac in several important ways. First, it was less dependent on information collected during exploratory behavior in the arms. Second, the operant sample phase allowed the animal to select a new sample each trial, while the delayed alternation T-maze task described above had one sample to start the entire set of trials, requiring that the animal alternate from arm to arm on each subsequent trial. However, similar to the delayed alternation task described above, the animal could adopt a non-working memory based strategy to solve the task. In this case, after selecting the initial sample lever, the animal could immediately wait in front of the opposite lever until it appeared, effectively overriding the need to utilize working memory during the delay time. Nevertheless, in this task, Rossi et al found that lesioning the mPFC of mice did impair performance. Additionally, in this paradigm, the animal had a fixed amount of time to perform as many trials as possible, and animals with an mPFC lesion performed significantly more trials and made significantly more lever presses. At one level, this contrasts with our findings, as we saw no effect of the mPFC lesion on the time to complete each trial or in the latencies between different actions within a trial. However, given that in the Rossi et al task, the animals could complete as many trials as possible in ninety minutes, the lesioned animals may have been able to attempt more trials than controls because their higher error rate resulted in less time spent consuming rewards. We would not be able to detect this in our experiment given that our daily sessions were based on a fixed number of trials rather than a fixed amount of time.

Altogether, this rich literature suggests that the mPFC is required for working memory as assayed in a variety of different T-maze and operant-based tasks. Our results support this notion by demonstrating the necessity of the region for this operant-based version of the DNMS paradigm.

The mPFC is not necessary for all operant-based cognitive tasks.

While many DNMS tasks, including the one presented in this study, are dependent on the mPFC, this is not necessarily the case for all operant-based tasks where information needs to be maintained over a sustained period of time. For instance, in a different operant-based task where animals need to press the opposite lever from that indicated by a visual cue following a delay, Kahn et al found no effect of a mPFC lesion (Kahn et al., 2012). This apparent conflict with our findings may be explained by several differences between the tasks.

First, in the task presented here, we introduced all delay lengths within a single day, a more complicated trial structure that required an adjustment period, whereas Kahn et al tested one delay length over three days before switching to the next delay length. Our data do not indicate that the impairment we see is due to a difference in learning during the delay stage; however, it is possible that the DNMS task is simply more difficult than the Kahn et al task and therefore requires the mPFC while the easier task did not.

Second, as with the task described in Rossi et al., in the task described in Kahn et al, the animal could adopt the alternate strategy of waiting in front of the correct lever immediately following the presentation of the light cue, negating the need to use any working memory. Given that the animals in Kahn et al were heavily trained on the task, having completed 6 weeks of training followed by a 3 week-long sustained attention test before beginning the delayed version of the task, it is possible that they adopted this more efficient strategy by the time the working memory test began.

Third, the difference between our results may arise because the cued information is passively given in the Kahn et al task, where our task requires the animal to move to press the cued lever. Several studies have shown that locomotion can enhance neural activity and learning in response to visual stimuli, indicating that the involvement of movement in the sample encoding phase of our task may lead to a difference in the cognitive processing of this phase of the trial (Dadarlat & Stryker, 2017; Kaneko, Fu, & Stryker, 2017; Pakan, Francioni, & Rochefort, 2018; Stryker, 2014). Thus, differences in the trial structure between the Kahn et al task and the one presented in this study may be quite substantial on a neural and circuit level, explaining the differences in mPFC dependence.

The mPFC is part of a larger network supporting working memory performance in this task.

Previous studies have used in vivo electrophysiology to investigate the roles and relationships of the hippocampus, MD, and mPFC during the different task phases of the DNMS T-maze paradigm (Abbas et al., 2018; Bolkan et al., 2017; Parnaudeau et al., 2013; Spellman et al., 2015). This approach has led to critical new insights into the properties of these circuits, including the directionality of essential projections during different task phases (e.g., vHPC-to-mPFC during the encoding phase, MD-to-mPFC during the maintenance phase, and mPFC-to-MD during the choice phase) and the differential roles of specific cell types involved (e.g., parvalbumin vs. somatostatin interneurons). In addition to the mPFC, the DNMS operant task used in this paper has previously been shown to rely on the dorsomedial striatum (Akhlaghpour et al., 2016) and the hippocampus (Goto & Ito, 2017; Goto et al., 2010). With all of the parallels between the T-maze and the operant-based DNMS tasks, we might expect the MD, which is crucial for the maintenance and choice phases of the T-maze, to play a similar role in this operant task.

Given the multi-region network that appears to be involved, an electrophysiological approach similar to what has been done with the T-maze might be used down the line to explore the communication between regions during the DNMS operant task. Some studies already have started to ask these questions in similar tasks (Hyman, Zilli, Paley, & Hasselmo, 2010), but the exact circuitry involved is still unknown. Furthermore, given the importance of mPFC in task acquisition, physiological recordings should also be done during this period. These studies would allow us to better understand the circuit connectivity involved in the encoding, maintenance, and selection of the correct lever. They might include: 1) in vivo electrophysiology to measure local field potentials in the mPFC, hippocampus, dorsomedial striatum, and MD to explore any directionality in coherence of oscillatory activity between the regions during the task; 2) acute optogenetic silencing or activation of specific interneuron populations in the mPFC to elucidate their differential roles; or 3) acute manipulation of projections between the mPFC and the MD, hippocampus, or dorsomedial striatum to understand the specific elements of circuit connectivity required for the task.

In sum, this study demonstrates the importance of the mPFC in an operant-based working memory task. This operant task has several advantages over T-maze DNMS tasks. First, the short task phase allows a large number of trials to be assessed each day, enabling multiple different conditions and manipulations to be introduced on a given day. Additionally, this short task phase allows for temporally precise manipulations and interpretations of task-related neural activity when combined with in vivo recordings and imaging studies. In addition, the automated nature of the task allows for multiple animals to run simultaneously in an experimenter-independent environment. With this task, future experiments will be able to delve deeper into questions regarding the neural circuitry and communication contributing to this type of cognitive, prefrontal-dependent behavior.