An Appetitive Spatial Working Memory Task for Mice in a Semi-Automated 8-Arm Radial Maze, Reducing Fearful Memory Association in the Maze

Summary

Most behavior experiments for rodents were originally devised for rats. Due to the behavioral differences between rats and mice, sometimes modifications are required for mice. We present a method to test spatial working memory in mice, minimizing the stress during the working memory task.

Abstract

This method paper details a protocol to test spatial working memory in mice using a semi-automated radial 8-arm maze (RAM). The RAM is a partially enclosed apparatus with 8 horizontal, equally spaced arms radiating from a central hub, from which access to each arm can be controlled individually by servo-controlled motorized doors. Animals start in the central hub and are allowed to explore the maze for a food reward at the end of each arm or selected arm. The RAM task was originally designed for rats, but we have adapted the protocol for mice, for example, by including more habituation steps. In our protocol, all arms are initially baited with sweetened condensed milk, and mice are admitted sequentially into four pseudo-randomly selected arms to collect the rewards ("forced run") before opening all doors together to allow the mice to run freely and find the remaining four rewards ("free run"). A 5 s delay is introduced between the forced and free runs to challenge working memory; an error is recorded if the mouse enters any previously visited arm during the free run. The task is complete when all rewards are recovered. After 6 days of habituation and 9 days of maze training, male C57BL/6 mice regularly achieve ≥ 80% daily success rate score, defined as 4/(4+E), where E is the number of errors. This semi-automated task could, in principle, be combined with in vivo monitoring methods such as electrophysiology, multiple-photon microscopy, or calcium imaging.

Introduction

Working memory¹ is a cognitive process that can hold a limited amount of information for a short period while it is manipulated and used in the planning and execution of a mental task, such as memorizing a phone number or a route for a destination from a different location. It is closely related to fluid intelligence, which is the ability to reason quickly and think abstractly. It is believed that working memory training can improve fluid intelligence². Working memory, therefore, is crucial for research studies.

Most popular behavior experiments for rodents were originally devised for rats in the 1980s, and researchers today often apply those protocols to mice. However, there are consistent behavioral differences between rats and mice^{3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19}. Especially, fear management is an important factor that can affect the results of the experiments, as mice are more prone to stress compared to rats^{3,4,5,6,7,10,11,12,13,14,15,16,17,18,19}. Therefore, it is important to consider the differences when applying those protocols to mice. In our protocol, we tried to minimize the stress factors during the working memory task in the radial arm maze. This was done by carefully timing the door closures using the mouse tracking and maze control software. Also, mice catching at the end of the session in the central hub was avoided. This approach allowed the mice to choose their next arm without associating the location with the fear of being caught. Delaying the capture of the mice until after all rewards had been collected in the maze ensured that the last reward was not linked to the experience of being caught.

The land radial arm maze (RAM) is a popular and commonly used apparatus to assess spatial and nonspatial learning and memory, and to test working and reference memory in rodents. It is one of the most popular apparatuses for spatial learning and memory in mice^20,21,22. RAM takes advantage of the natural tendency of rodents to explore the maze, find novelties, and learn spatial information on a trial-and-error basis, in addition to foraging. This tendency can be augmented by restricting their diet and rewarding them with preferred food in the maze. Nowadays, several variants of the RAM apparatus are used, e.g., with 6, 8, 12, or 16 arms and a partially enclosed or open/elevated apparatus^23,24,25. Most present-day protocols are based on that of Olton et al.^20,26, but use simpler (6- or 8-arm) mazes^{9,24,27,28,29}. Testing working memory in a radial arm maze (6 or 8 arms) is more powerful than the T maze (2 arms) because animals have more options of arms to explore.

The original RAM was designed by Olton and Samuelson²⁰ to assess spatial learning and memory in rats using food baits for positive reinforcement, unlike water mazes such as the Morris water maze and 8-arm radial water maze that use negative reinforcement, i.e., water immersion^20,30,31,32. Olton and Samuelson²⁰found that rats have excellent memories of places as they pass through the RAM. They showed that the hippocampus is involved in spatial information processing. They, therefore, introduced the concept of a dual memory system, comprising working and reference memory, in rodents. This dual memory system had been suggested previously by Honig³⁶  based on experiments with pigeons. Olton et al. found that hippocampal function is essential for working memory. This was achieved by making lesions in the entorhinal area, the body of the fimbria-fornix, anterior to the hippocampus, septum, or postcommissural fornix to disrupt each of the major extrinsic fiber connections of the hippocampal formation^26,33,34,35. Fornix-fimbria lesions did not affect reference memory²⁶. 

In modern protocols, animals are habituated to the environment by allowing them to explore the maze freely for 10-15 min per day for 3 consecutive days, followed by maze training for 1-3 weeks with a preferred food at the end of all arms or fixed selected arms, testing working memory, or working memory and reference memory in the same experiment, respectively^9,29,37.

Often, these protocols are not described fully, and the behavioral differences between rats and mice are not well considered. Here we present our protocol and experience in detail, especially how to avoid associating fearful experiences with places and timing based on fear management differences between rats and mice. This protocol is for researchers who want to study spatial working memory and apply it to electrophysiology, real-time imaging, and histology using mice.

Protocol

All mouse experiments were pre-approved by the UCL Ethical Committee and authorized under the Animals (Scientific Procedures) Act 1986 and subsequent legislation of the UK Government.

1. Preparation of programs for the automated maze control on a software application

NOTE: The automated radial arm maze, as shown in Figure 1A-C, and the room layout in Figure 1D were used.

1. Prepare an experimental workflow for door closure as described in Figure 2.
2. Set the start of the habituation stage 1 (Figure 2A) on a software application. To do so, follow the steps below.
   1. Set the start of the trial and video recordings, and close all the arms on the software. 5 s after the software recognizes the mouse in the maze's central hub, it starts tracking and opens all the doors.
   2. When the mouse goes inside an arm and is 4 cm away from the closest door of the central hub doors, close all the doors. Then, immediately after closing all the doors, open the door of the arm that the mouse selected.
   3. Once the mouse returns to the central hub, close the door and confine it to the central hub for 5 s. Repeat this for 30 min.
   4. Stop the video recording and tracking when 30 min elapse, and the mouse returns to the central hub.
      NOTE: Let the mouse explore the maze freely at their will.
3. Habituation stage 2 (Figure 2B)
   1. Set the start of the trial and video recording in the associated software and close all the doors.
   2. 5 s after the software recognizes the mouse in the central hub of the maze, start its tracking and open all doors.
   3. End the trial if the mouse enters each arm at least twice or after 30 min and then returns to the central hub. Let the mouse explore the maze freely at its will.
4. Habituation stage 3 (Figure 2C)
   NOTE: This stage is the combination of stages 1 and 2.
   1. Set the software to start the trial and video recording and close all the doors.
   2. Set the software to control the maze doors 5 s after the software recognizes the mouse in the central hub. Set the start of the tracking at this point and open all the doors.
   3. Close all the doors when the mouse goes inside an arm and is 4 cm away from the closest door of the central hub doors. Then open the door of the arm.
   4. Close the door when the mouse fully enters the central hub and confine the mouse in the central hub for 3 s. Repeat this sequence for 30 min or until the mouse enters each arm at least once.
   5. Set the end of the trial after fulfilling the condition, and when the mouse enters the central hub fully.
      NOTE: Let the mouse explore the maze freely at its will.
5. Working memory task (Figure 2D, E)
   1. Set the software to start the trial and recording, and close all the doors.
   2. 5 s after the software recognizes the mouse in the central hub, set the software to start the tracking.
   3. Start the forced run phase. Present pseudo-randomly chosen 4 arms one by one to the mouse.
   4. When the mouse enters the first arm and is 3 cm away from the central hub, ensure the door closes and confines it in the arm for 15 s.
   5. Let the door open and let the mouse go back to the central hub.
   6. Close the door and then open the second arm.
   7. Repeat this till the fourth arm.
   8. When the mouse returns to the central hub after the fourth arm of the forced run phase, confine it to the central hub for 5 s.
   9. Let all the doors open and start the free run phase.
   10. End the trial when the mouse returns to the central hub after collecting all the rewards/entering all the arms at least once.

2. Preparing home cages for individual housing

1. Set up new cages using standard nesting materials for a new experiment.
   1. Put new bedding, sawdust, a paper igloo, a paper tunnel, paper nesting stripes, and a wooden bite block in new cages for each mouse for the experiment.
   2. Place the food hopper and close the lid.
      NOTE: Mice can damage the parts for ventilation if they are without the food hopper.

3. Handling, weighing, and diet restriction in the animal holding room (Day -7 to Day 0)

1. Guide the mouse to the paper tunnel supplied in the home cage by hand gently.
2. Lift the mouse from the grouped cage using the paper tunnel^38,39,40 with the hand, keep it at chest height for several seconds. Alternatively, let the mouse climb onto the hand voluntarily.
   NOTE: Handling animals with minimal stress is essential for behavior experiments. Never pick them up by the tail, which is stressful for mice. Habituating mice to the experimenter is also important. Mice need to be well-habituated and relaxed, or they will not learn well in the maze later.
3. Always hold the mouse using the same handling method. Put the mouse on the scale for body weight monitoring and record its body weight.
   NOTE: Do not stress the mice with fast body movement or by making loud noises. It is better to do this in a quiet environment. Otherwise, they may jump off the scale or the experimenter's hand. Also, if the mice are stressed, they can show squinted eyes, squeak, and may bite the experimenter's hand^41,42.
4. After weighing, put the mouse into the new cage prepared earlier.
5. Put mice on a diet with a restricted amount for a week prior to the exposure to the radial arm maze. This motivates them to forage the maze for food rewards provided. The food reward during the experiment is diluted sweetened condensed milk.
6. Put 2 ~ 3 g of standard chow in the cage and maintain the body weight between 85 ~ 90% of the starting weight throughout the experiment.
   NOTE: C57BL/6 mice eat 3-4 g of standard chow every day⁴³.
7. Observe the overnight drop in the body weight as a guide to standardize the amount of food to be given the following day. From then on, give a sufficient daily amount of standard chow to maintain 85-90% of their starting body weight throughout the experiment.
   NOTE: These animals need to lose weight slowly. Monitor their body weight and facial expressions to avoid stress.
8. Provide 1 mL of diluted condensed milk (reward; sweetened condensed milk : water = 1 : 1) in a small plastic cup in the home cage to habituate mice until the end of the habituation stage 1. Make sure mice eat the reward. Change the reward food if the mice strain shows neophobia to the presented food.
   NOTE: From habituation stage 2, do not provide the reward in the home cage. If the body weight of any mouse drops below 80% of the starting weight, it should be euthanized as per the IACUC and country-specific guidelines. If a mouse is not in the normal body weight range for the mouse line being used (20-30 g at 8-12 weeks old) at the first body weight check, it should not be considered in the cohort to avoid difficulty in maintaining its body weight in the required range later.

4. Bringing mice to the testing room

1. Use a trolley to bring mice to the behavior testing room slowly.
2. Leave the mice undisturbed on the trolley for 5 min.

5. Room set up

1. Set the room lighting illuminating the central hub indirectly, at 80 lux with warm white light bulbs, and play white noise at 70 dB measured at the central hub. NOTE: It is known that exposure to moderate-intensity white noise leads rats to perform better in a geometrically complex maze^44,45. White noise seems to do more than simply mask the unavoidable noise in the environment; 70 dB white noise increases choline uptake in several brain regions of rats, including the prefrontal cortex, hypothalamus, and hippocampus (but not in the striatum)⁴⁶ and RAM working memory performance is known to depend on the integrity of cholinergic pathways among the hippocampus, frontal cortex, and forebrain^32,47. It is also known that animals use both distal and proximal visual cues to navigate accurately in their environments. However, mice depend on distal visual cues more than proximal cues to navigate during spatial learning⁴⁸. In this experiment, proximal cues printed on a small piece of paper were introduced to the back walls of all the arms of the 8-arm RAM, but they did not reduce the daisy-chaining behavior. Geometrically asymmetrical distal visual cues (on the ceiling and walls of the room), on the other hand, reduced daisy-chaining and were essential for spatial navigation in the RAM.

6. Habituation stage 1 - Non-rewarded (Day 1 - 2)

NOTE: No rewards are baited in the maze. This stage is to habituate mice to the maze doors and their sounds.

1. Start the mouse tracking and video recording by clicking the Start button on the software and bringing the mouse to the maze without stressing it. Do this by moving slowly and changing the height slowly while the mouse is in the experimenter's hand. Place the mouse in the central hub.
   NOTE: If mice shake on the experimenter's hand, slow down the movements. It is a sign of being scared. Mice move around during the 5 s confinement period in the central hub before starting the forced run phase. They can face any side when put in the central hub.
2. Keep watching the mouse's behavior on the PC monitor placed far from the maze or below the maze wall level. If the mouse tries to escape from the maze by climbing on the wall, simply stand up from the chair to show the experimenter's existence and walk toward it, monitoring their perfomance. In the worst-case scenario, push the mouse back into the maze.
   NOTE: Normally, mice go back to the maze after hearing the frictional noise of the experimenter's clothes as they approach it. This needs to be done quickly. If the experimenter continues doing this, the mice will stay in the maze after the habituation stages 1 through 3. If mice keep jumping out of the maze, it is advisable to remove them from further experiments and analysis.
3. After the trial, collect the mouse from an arm of the maze by letting the mouse climb on the experimenter's hand voluntarily while the other hand covers the arm to prevent it from jumping off the maze.
4. Brush off mouse feces and wipe the maze with dry tissue to remove any traces of the rewards. Then, wipe with wet tissue and dry the maze with dry tissue (about 2 min).
5. Do this for each mouse in turn for two days.
   NOTE: After 30 min, mice might not have visited all 8 arms of the maze. This is normal and is not a problem. Mice tend to visit arms in an unsystematic, non-sequential pattern and move cautiously, spending uneven times in each arm. Mice can climb on the maze walls and/or jump out of the maze at this stage. If this happens (e.g., the mouse tries to climb on the walls), prevent it by ensuring that the experimenter makes the mouse aware of his/her presence by approaching the maze and making some noise, such as clothes rustling. Therefore, careful observation of the mouse's behavior in the RAM and quick action might be required. After a few repetitions of this procedure, mice will stay in the maze. If, despite everything, the mouse does not stay in the RAM during habituation, remove it from the analysis.

7. Habituation stage 2 - Rewarded (Day 3 - 4)

NOTE: This stage is to encourage mice to explore the maze freely for food rewards.

1. Place 70 μL of the food reward in each food well in each arm.
2. Start the mouse tracking and video recording by clicking the Start button on the software and then placing the mouse onto the maze central hub as explained in 6.1.
3. Keep watching the mouse's behavior on the PC monitor from a distant place and write it down if necessary, e.g., climbing on the wall or jumping off the maze. End the trial when the mouse enters each arm at least twice or 30 min have elapsed.
4. After the trial, collect the mouse from the maze in an arm by letting the mouse climb on the experimenter's hand voluntarily while the other hand covers the arm to prevent it from jumping off the maze.
5. Clean the maze as described in 6.4 (Habituation stage 1; about 2 min).
6. Do this for each mouse in turn twice per day for two days.
   NOTE: At this stage, mice may display daisy-chaining behavior, but it is unnecessary to worry because the additional 4 arm entries in the working memory task may discourage this movement. In this step, mice can stop in the arm without consuming the reward. On Day 4, mice should be more trained than on Day 3 but can still stop in the middle of the arm.

8. Habituation stage 3 - Rewarded (Day 5 - 6)

NOTE: This stage is to let mice get used to the environment of the combination of stages 1 and 2.

1. Bait all the food wells with 70 μL of the reward.
2. Start the mouse tracking and video recording by clicking the software's start button and placing the mouse at the maze's central hub.
3. Keep watching the mouse's behavior on the PC monitor from a distant place, until the mouse enters each arm at least once or 30 min elapse. If the mouse tries to escape from the maze, act fast as stated in Habituation Stage 1 (6.2).
4. After the trial, remove the mouse from the maze by letting it climb on the experimenter's hand voluntarily while the other hand covers the arm to prevent it from jumping off the maze.
5. Clean the maze as described in 6.4 (habituation stage 1) (about 2 min).
6. Do this for each mouse twice per day for two days.
   NOTE: At this stage, removing any mice attempting to escape from the maze from the analysis is best.

9. Working memory task - Rewarded (Day 7 - Day 15)

1. Bait all the arms with 40 μL of the reward. Place the reward in each food well close to the central hub to make it difficult for mice to see from the central hub.
2. Start the mouse tracking and recording by clicking the software's start button and then place the mouse onto the maze central hub without stressing it.
3. During the 5 s confinement in the central hub, walk away and sit somewhere away from the maze. Monitor the mouse's behavior and write down the sequence of the arm that the mouse visits in the maze (see Supplementary file 1).
4. Stop recording the arm visit sequence when the mouse collects all eight food rewards, which is the end of the working memory task, and then let it visit four extra arms.
   NOTE: By doing this, mice can have some more time before being caught, which allows mice to focus on the working memory task. This step also allows mice to reinforce the fact that rewards are not replenished once collected.
5. After the four extra arm visits, remove the mouse from the maze at the end of the last arm visit by letting it climb onto the hand while covering the arm with the other hand.
   NOTE: By not catching mice in the central hub, mice will not associate being caught with the central hub, where mice decide which arm to go to next. Once mice become familiar with the task, including the extra arm visits, they can freely move near the last food well or spend some time in the central hub. However, they should finish the 4 extra arm visits within 3 min. If they do not finish it within 5 min, return them to the home cage.
6. Select the next cage to adjust the arousal level by tapping the cage for the next trial.
7. Clean the maze as described in 6.4 (habituation stage 1) (about 2 min).
8. Move on to the next mouse and repeat this 6 times for each mouse in turn for 9 days.
   NOTE: The average duration of the four extra arm visits should become longer in later trials, as mice learn to recognize when no further rewards are available or that there are no novel, unvisited arms to go into.

10. Scoring

1. Calculate "daily success rate score" (%) = 4(goals) / (total arm entry number[4+Errors]) x 100.
   NOTE: After the forced run, only four arms are filled with the reward, the goals. If they enter an arm the mouse visited in the forced run, count it as one error, as well as repeated arms in the free run.
2. Calculate "perfect trial ratio score" (%) in 6 trials = [(number of perfect trials) / 6 ] x 100
   NOTE: A perfect score means there is no working memory error.
3. Select speed, distance, and acceleration of the mouse in the software for analysis and calculate the averages for each day. Ensure to exclude the four extra arm visits from these analyses. Also, measure the duration of the four extra arm visits by reviewing the mouse tracking videos after the experiment.
   NOTE: Mice can make quick turns immediately after entering an arm. Do not count and score this as an entry, as it is a quick mistake correction, which means working memory is valid⁴⁹. However, if the mouse crosses the midpoint of the arm or stays in the arm for more than 3 s, count and score the entry. Daily success rates scores indicate mice's learning curves, and perfect trial ratio scores indicate working memory retention efficiency.

11. Statistics

1. If studying the difference between the control and a knockout mouse line, as done in the previous publication⁵⁰, perform the statistics as described below.
   1. Compare the resulting data of the control and experimental groups using repeated measures two-way ANOVA with a post hoc analysis (e.g., Bonferroni test) [measuring each subject at 9 different time points].
   2. If comparing groups, test at least 10 and preferably ~20 mice in each group. Where the effect of gene manipulation, making lesions, or drug treatment is obvious, the sample sizes can be smaller to conclude.

Results

This protocol was used in our previous publication, where we asked whether new oligodendrocytes and myelin sheaths are necessary for spatial working memory task performance using MyRF cKO mice and the control⁵⁰. We found that the working memory task performance was improved in the control mice but not in MyRF cKO mice. The working memory task stimulated and promoted oligodendrocyte precursor cell (OPCs) proliferation and differentiation, and oligodendrocyte (OL) myelination in the brain areas related to working memory processing in wild-type mice. We also performed the T-maze rewarded alternation task to assess spatial working memory. This task was performed to detect working memory deficiency in MyRF cKO mice. However, no significant differences were observed in the histological analysis between the mouse group that had the T-maze rewarded alternation task and the home cage control group. Therefore, we concluded that the protocol presented here is more effective than the T-maze rewarded alternation task for studying working memory.

Each mouse had 6 trials of the working memory task per day that lasted for one hour in total. Thus, this protocol allowed for a maximum of 8 mice in a cohort. In the early days of habituation, mice spent uneven time in the maze arms, and their movements were slow. They stayed at the end of the arms for a long time. Until the end of day 5 of the habituation in the maze, they spent variable times in the maze arms and could return to the central hub without touching the rewards (Figure 3A). However, on the sixth day, animals entered each arm at least once within 8-15 min, spent a similar period in each arm, which indicated that the animals were habituated to the maze and were ready for the task (Figure 3A, day 6 lower row and Figure 3B).

During the habituation stages, they often stood on their hind limbs and looked up from the arms or the central hub. Similar behavior was observed during the working memory task, but they did not necessarily go around in the same direction (e.g., 2 -> 3-> 5 -> 8 [clockwise] or 8 -> 5 -> 3 ->2 [anti-clockwise]). They could change their direction (e.g., 5 -> 4 -> 6 -> 1). Individual mice had no apparent communal tendency to choose the arm sequence. Mice tend to spend a little more time selecting the last rewarded arm than the first rewarded arm in the free run. This was probably because they had four options to choose from at the first choice in the free run, and only one option for the last reward. The first arm in the forced run phase was presented to mice at least 50 (15+15+15+5) s before the start of free run. Thus, this working memory task was not very easy to perform without working memory errors. Supplementary Video 1 shows a mouse undergoing a perfect trial on day 8 of the working memory task training.

Using either scoring system, we found that their performance was almost flat for the first 3 days of the 9-day RAM testing period but improved markedly after that (Figure 4A, B). After 6 days of habituation and 9 days of maze training, male C57BL/6 mice regularly achieved ≥ 80 % daily success rate score from day 8, defined as 4/(4+E), where E is the number of errors. The average distance traveled in the free runs became shorter over time, but the average speed and acceleration were fairly constant from day to day (Figure 4C-F). Mice might have developed a fear memory of being caught during the four additional arm visits, which could be one of the reasons why mice spent more time towards the end of the experiment (Figure 4G).

In this protocol, habituation steps lasted for 6 days, which could lead to adjusting the body clock for the experiment, even if the light/dark cycle was not reversed in an animal facility.

Figure 1: RAM specifications and features. (A) Views of the maze from the ceiling-mounted CCD and from the side, and the dimensions. (B) The food well was sunk 1.5 cm below the maze floor level so that mice cannot see the contents from >10 cm away. (C) Maze arena setting. The central hub and arm 1-8 are shown. (D) Schematic of the experimental room and the external visual cues with a photograph of the room. Please click here to view a larger version of this figure.

Figure 2: Flow diagrams of the maze control programs. (A) Habituation stage 1 (days 1 and 2). (B) Habituation stage 2 (days 3 and 4). (C) Habituation stage 3 (days 5 and 6). (D) Working memory task (days 7-15). (E) Schema of the working memory task paradigm. Mice went through the pseudo-randomly chosen four forced run arms, were confined in the central hub for 5 s, and then went for the free run to collect the remaining rewards. Please click here to view a larger version of this figure.

Figure 3: Heat maps and bird nest maps during the 6-day habituation. (A) Heat maps of mouse trajectories during habituation days 1-6, for three mice. For habituation days 3-6, mice underwent two trials per day; the upper heatmaps show the first trials and the lower show the second trials. (B) Bird nest maps for habituation day 6. The upper three show the first trials and the lower three show the second trials of mice 1, 2, and 3. The lower Table lists trial durations for the same three mice. Please click here to view a larger version of this figure.

Figure 4: The RAM working memory task. (A) Average daily success rate scores (%) calculated by [4/(number of arm entries during the free run) x 100]. Scores from 6 trials were averaged for the day. (B) Perfect trial scores over the 9 days of the task. (C) Individual distances traveled and the average. (D) Individual running speeds of the mice in the task and the average. (E) Individual accelerations of the mice and the average. (F) Individual decelerations of the mice and the average. (G) Individual and average durations of the additional 4 arm visits. Please click here to view a larger version of this figure.

Supplementary video 1: The mouse undergoes a perfect trial on day 8 of the working memory task training. Please click here to download this File.

Supplementary File 1: Example recording sheet. Write down the sequence of the arm visits while watching the mouse's behavior. Please click here to download this File.

Discussion

Training and testing working memory in the RAM
Working memory is a cognitive process that can hold a limited amount of information for a short time while it is manipulated and used in the planning and execution of a mental task. Examples include memorizing the entry code for a door at a remote location or memorizing the path to a specific destination. The dual N-back task and spatial span task are commonly used visuospatial working memory tasks in which human subjects are asked to memorize the location, identity, and temporal order of the stimuli. Takeuchi et al. found that the amount of training in successful working memory tasks in humans, which include the dual N-back task, correlates with the increased fractional anisotropy in the anterior part of the body of the corpus callosum in their study using MRI⁵¹. In our earlier study⁵⁰, using the protocol described in this paper, we found that myelinating oligodendrocytes increased in the anterior corpus callosum and other brain areas that are thought to be involved in working memory processing, including the anterior cingulate cortex, which is believed to be involved in attention. The oligodendrocyte increase closely correlates with their working memory task performance, which is consistent with the study in humans by Takeuchi et al^50,51,52. However, it is still necessary to study whether there are similarities in psychological processes between humans and mice (e.g., electrophysiology). Working memory is closely related to attention. It correlates with measures of 'fluid intelligence' in humans and is important for learning, hence important to study. Olton and Samuelson described an 8-arm radial maze (RAM) task for training and assessing spatial memory in rats²⁰. In the free choice experiment, they provided all maze arms with 0.1 g rat chow, which was not replenished during the test. So, only eight rewards were obtained in one test. All arms were open, and rats had immediate access to all the arms during the experiment. Since food rewards were not replenished, rats had to use working memory to avoid revisiting previously visited arms, in order to recover as many rewards as possible before the end of the task. Each rat was placed in the central hub at the start and allowed to enter any arm at will, to recover the food reward until 16 arm choices were made or 10 min elapsed. Olton et al. counted the number of distinct arms rats visited within the first 8 choices and found all rats chose an average of more than 7 distinct arms within 10 days²⁰. We tried a similar protocol with mice and found that mice, unlike rats, showed daisy-chaining for a few days and did not change their strategy afterward, which circumvented the purpose of assessing their working memory. The mouse was, therefore, confined in the central hub between arm visits, hoping confinement in the central hub would prevent the daisy-chaining behavior.

Olton et al. confined rats in the central hub for 1 min or 15 s to interfere with any sequence of choices. This prevented them from adopting their usual tendency to progress around the maze continuously, in a clockwise or anti-clockwise direction^20,53. If the choice of any arm is random, equally probable, and completely independent of previous choices, the experiment is an example of the classic 'occupancy problem'²⁰, so the probabilities p(n,8) of selecting n distinct arms within the first eight successive arm visits are: p(1,8)=4.8x10^-7, p(2,8)=4.2x10^-4, p(3,8)=0.019, p(4,8)=0.17, p(5,8)=0.42, p(6,8)=0.32, p(7,8)=0.067, p(8,8)=0.0024. From these probabilities, one can calculate that the mean chance likelihood of choosing distinct arms is 5.3 (Expectation value = = 5.25; 5.3/8= 66%) out of 8. Olton et al. reported that, over a 10-day training period, the rats entered on average 5.7/8 distinct arms (71%; range: 4.2-6.6) over days 1-5, increasing to 7.6/ 8 distinct arms (95%; range: 7.2-8.0) during days 6-10²⁰, which is significantly higher than chance. This indicated that rats used working memory to avoid food-depleted arms. We tried a similar approach with mice, confining them to the central hub for 5 s between arm visits. Mice entered an average of 5.7/ 8 different arms per 8-arm trial over days 1-5 (71%; range: 5.2-5.8; eleven C57BL/6 mice) and 5.9/ 8 arms during days 6-10 (74%; range: 5.7-6.5; eleven C57BL/6 mice). Therefore, mice were unable to learn, or learned very slowly, to improve their reward-finding performance significantly above chance levels during the 10 days of the experiment. Therefore, the protocols above were considered as unsuitable for mice, opting instead for a 'forced run/free run' paradigm, similar to that described for rats by Sasaki et al.²⁷

In the protocol of Sasaki et al.²⁷, all arms were baited at the beginning of the experiment, rewards were not replenished, and each rat was initially confined to the central hub of an open elevated 8-arm RAM. They were then released into each of the 4 pre-selected arms, one arm at a time, without confining them in the arms and the central hub between arm visits (the 'forced run'). Following the 4^th arm visit, the rat would go back to the central hub, and all doors were immediately opened to allow the rat to visit any and all arms at will (the 'free run'). In order to collect all the remaining food rewards in the most efficient way during the free run, animals need to remember and avoid any arms they have already visited in either the forced or free runs. We used a very similar approach (Figure 2E) with mice. Ten minutes of habituation periods at the beginning of every day's RAM training in Sasaki et al.²⁷ was replaced with a single 6-day habituation period prior to the training. Another modification made was allowing mice to make four additional arm visits at the end of each trial once all the rewards had been recovered. This was intended to reinforce the fact that rewards are not replenished in previously visited arms. Also, since mice learn to "avoid regret"^54,55, it might discourage daisy-chaining behavior in the following trial, because such behavior would go unrewarded during the four additional arm visits immediately prior to their starting the subsequent trial. Another reason for the additional four arm visits was to prevent the mice from associating a successful recovery of all rewards with the possibly fearful experience of being caught by the experimenter at the end of the task. Mice might have developed a fear-memory when caught during the four additional arm visits, which could be one of the reasons why they spent more time towards the end of the experiment (Figure 4G). Indeed, mice broke the daisy chaining behavior during the additional four arm visits, if they did it in the trial. Furthermore, mice were confined in each arm for 15 s during the forced run and then in the central hub for 5 s before opening all the doors to start the free run. Without confining the mouse in the central hub, mice tended to go daisy chaining as described above.

For the forced run, we selected from a set of five 4-arm pseudorandom sequences that contain 3 or fewer adjacent arms. These were (1,3,5,7), (2,4,5,8), (1,4,5,7), (2,3,4,7), (1,2,5,6), and their variations of different arm orders and rotations, numbering the arms from 1-8 clockwise from the top (Figure 1C). Each mouse in a cohort underwent six trials (one trial = forced run +5 s confinement in the central hub + free run) per day during the 9-day post-habituation stage of the RAM test. During the free run, if a mouse made only four arm visits to recover all four rewards remaining after the forced run, it was termed a "perfect trial". Their performance during the 9 days of training was quantified by "daily success rate score", calculated as 4/(4+E), where E is the number of errors during the free run or the proportion of trials that were "perfect trials" (i.e. E=0). For example, if a mouse needed 7 arm visits to retrieve the 4 remaining rewards during the free run then the "daily success rate score" would be 4/7 = 0.57 or 57% and a perfect trial would score 100% on this scale; if a mouse made 5 perfect trials on the day the perfect trial ratio score would be 5/6 = 83.3 %.

Daily success rate scores achieved by daisy chaining were manually calculated for each of the 4-arm forced run patterns that were employed (previous paragraph), assuming that mice start their free run at random in any one of the 8 arms of the RAM. The average score over many trials that can be achieved by daisy-chaining is between 53% and 56% for the individual forced run arm patterns, or ~55% overall arm patterns. The probability of obtaining a perfect score (100%) by daisy-chaining is zero, and that by random arm visits is <1% (4/8 x 3/7 x 2/7 x 1/7 = 0.87%), assuming mice never return to the arm they have just exited. Therefore, the average daily success rate scores (>80%) and percent of perfect trial scores (>60%) reached by our mice in the final 2 days of the 9-day testing period (Figure 4A,B) could only be achieved consistently using a working memory-based strategy.

Mice that were familiar with the task tended to loiter at the last food well, or in the central hub, once they had consumed all the rewards, and they also tended to spend more time on the extra four arm visits at the end of the task (Figure 4G), suggesting that they had learned the "rule" that each arm contained one and only one reward, and also that they could keep track of the fact that they had visited all available arms. For mice of a different genetic background than that used here (C57BL/6), results might vary because genetic background matters in behavioral experiments^56,57,58. However, in our previous study⁵⁰, using mice with a mixed C57BL/6, CBA, 129P2 background but backcrossed with C57BL/6J mice, it was found that their working memory task performance was greatly improved using the protocol described here, with fewer mice resorting to daisy-chaining during the free run. Koike et al.⁵⁹ reported that all 129 mouse inbred substrains carry a Disc-1 deletion polymorphism consisting of a 25 bp deletion in exon 6 of Disc-1 that causes a frameshift and truncation of the DISC-1 protein, affecting its functionality^60,61. This mutation caused working memory impairment when studied with T-maze delayed alternation task⁵⁹, another appetitively motivated win-shift maze task, possibly explaining why the control mice used in our previous study⁵⁰ had slightly lower working memory scores than the inbred C57BL/6 mice. We always use mice that fall within the normal body weight range for their age and strain to avoid difficulties in maintaining their weight between 85% and 90%. Litter size could influence spatial memory and anxiety-like behavior because of early life overfeeding in a strain called NMRI⁶². Small litter size causes over-feeding and increased body weights in both the NMRI and C57BL/6 mouse strains. However, the two strains solved spatial learning tasks differently. While C57BL/6 mice did not show differences between small and control litter size, NMRI mice showed significant impairment in the learning phase with Morris water maze (MWM) test and slight impairment with Y-maze spontaneous alternation task though it was not significant. Furthermore, small litter size significantly decreased the protein levels of hippocampal brain-derived neurotrophic factor(BDNF) and increased the protein content of hippocampal interleukin-1b (IL-1b) after MWM and Y-maze tests in the NMRI strain. However, these changes were not observed in C57BL/6 strain. The NMRI strain of mice exhibited a significant increase in anxiety-like behaviors and stress-induced corticosterone levels when raised in small litter sizes. In contrast, small litter sizes for the C57BL/6 strain led to a reduction in both anxiety-like behaviors and corticosterone levels. Therefore, it's important to consider the litter size of the mouse strain when designing behavior experiments.

Behavioral differences between rats and mice
Most cognitive tasks were originally designed for rats. Also, spatial learning/memory has been extensively studied in rats. Because of the advantage of today's transgenic and genome modification technologies for mice, researchers often apply protocols devised for rats to mice. However, there are consistent behavioral differences between rats and mice^3,4,5,6,7. For example, they have preference differences (social preference by rats and food preference by mice)³ and use different strategies to negotiate different maze tasks (water or land mazes)⁴. Also, mice tend to take a longer time to achieve good spatial learning scores in 8-arm RAM^4,8,9, which is one reason that mice are generally regarded as inferior to rats in spatial learning tasks.

Rats and mice also differ in general behavior/character, including their responses to stress or fear management^3,4,5,6,7,10. For example, previous studies found poor fear extinction learning in adolescent mice^11,12,13,14, but not in rats^15,16,17,18. In addition, mice are more prone to stress and do not like being handled when compared to rats, which, when familiar with handling, apparently enjoy being tickled by the handler and try to interact with handlers for more tickling¹⁹. Therefore, fear management is an important factor to consider when studying mouse behavior. Furthermore, Borrow et al.⁶³ have reported that chronic variable stress decreases oxytocin mRNA, and increases the soma volume of oxytocin neurons in the paraventricular nucleus. Additionally, recent studies have shown that tunnel handling can reduce stress and improve performance in behavioral experiments^38,39,40. Therefore, it is important to avoid stressing mice during or before behavioral experiments.

Mice do not like to be exposed in the open, for example, in the open field test or elevated-plus maze, preferring more enclosed and darker places^64,65. In the context of the RAM, the central hub is the most open space, and it is from there that mice decide which arm to visit next. If they are anxious in the central hub, they might be distracted from the task and will perform below par. For this reason, we decided to catch mice in one of the radial arms (see procedure part), not in the central hub, so that they do not associate the hub with a fearful experience.

When exploring the 8-arm RAM for food rewards, rats emerging from a given arm tend to turn at right angles for their next arm visit, skipping the intervening arm, and rarely visit arms sequentially (what we term "daisy-chaining"); that is to say, rats tend to visit in the order 1,3,5,7 not 1,2,3,4 ^20,53,66. In contrast, mice often adopt a daisy-chaining strategy, possibly because of their body size and flexibility. The daisy-chaining strategy is a fairly cost-effective way for mice to recover the food rewards, but it is a considerable drawback for the experimenter trying to test spatial working memory because it circumvents the purpose of the experiment⁶⁷. However, stress exposure can affect learning ability and change learning strategy⁶⁸. Since the daisy-chaining approach may indicate stress, it is important to ensure that the mice remain unstressed. We describe our current protocol as stress-free for mice during the working memory task because mice do not use the daisy-chaining strategy.

It is known that mice rely on distal visual cues more than proximal visual cues when navigating the maze^48,69,70. We found that incorporating additional proximal visual cues that were printed on pieces of paper and placed on the back walls of all arms did not reduce such behavior, which is consistent with the claim by Olton et al. for rats²⁰. However, placing beacons at the goal arms as an intramaze cue helped mice more than extramaze cues in solving the task of the 6-arm radial maze reference memory and working memory testing in one experiment⁷¹. Furthermore, different mouse strains show different visual abilities⁷², but their visuo-spatial learning ability and memory tested in the Barnes maze did not necessarily reflect their visual ability, suggesting that mice with poor visual abilities, such as BALB/c, may rely on different brain processes. Thus, experimenters need to consider what cues they want to use to fit their experiments best.

Reducing anxiety in mice during handling
Handling stress is well known to influence animals physiologically and behaviourally^{73,74,75,76,77}. Also, Sandi et al demonstrated that spatial memory impairment caused by acute stress is associated with decreased expression of neural cell adhesion molecule (NCAM) in the hippocampus and prefrontal cortex by testing rats in a hippocampus-dependent water maze task using a female cat as a stressor between the last learning trial and the probe test⁷⁸, suggesting that the hippocampus and prefrontal cortex are vulnerable to high levels of stress. Therefore, it is important to reduce stress/anxiety in experimental animals to avoid unexplained variations within and between animal studies. Especially, working memory tasks require animals to focus during the task, so unnecessary distractions must be avoided. Published protocols for testing working memory in rats, or working memory and reference memory together, often do not include important details such as how, when, and from where experimenters remove the animals from the maze at the end of the task. Since mice are less social than rats with the handler and are more prone to stress, such details are more important for experiments with mice. Rodents can develop episodic memory, which is a form of explicit long-term memory that includes memory of times, locations, and associated emotions^79,80,81. Therefore, we chose not to remove mice from the maze immediately after they had collected all food rewards in the RAM but instead allowed them to make four subsequent arm visits without gaining food rewards. This was done to prevent the mice from associating successful recovery of all rewards with the possibly fearful experience of being caught by the experimenter, which is important, especially for the early stage of learning in the RAM. In addition, the additional four unrewarded arm visits can reinforce the fact that food rewards are not replenished in arms that the mice have previously visited. If the handling is not too stressful for mice, they interact with the experimenter more often³⁸. To avoid stress, lift the mouse from its home cage at the start of the RAM experiment using a paper tube in the home cage supplied when they were separated from the grouped cage for experiment³⁸, or else allow the mouse to climb onto the experimenter's hand voluntarily, when handling them in their home cage before the experiment and when bringing them from their home cage to the maze.When retrieving the mice from the maze at the end of an experiment, do not chase them in the maze but instead block the maze arm entrance with one hand and let the mice climb on it voluntarily, while covering the maze arm with the other hand to prevent the mouse from jumping out of the maze. This reduces the stress of being pursued in the maze or caught by the tail. Male experimenters are sometimes more stress-inducing than female experimenters, where mice are concerned⁸². However, by paying attention to the stress-reducing steps described here, it was found that male experimenters could conduct the RAM tests without major problems. Thus, how to handle mice is more important.

Sex differences
Only male mice are used in the experiments reported here because of the various impacts of sex in cognition, learning, and memory observed in different species (rodents and humans), and because the stress-coping responses differ between males and females^{63,83,84,85,86,87}. However, when interpreting results obtained with rodents for possible benefits for humans, it is important to study both sexes. Guidelines from NIH and other funding councils also suggest using both male and female animals.

Advantages of the protocol
Our semi-automated 8-arm radial maze is an enclosed maze with opaque white acrylic walls but no lids (see Figure 1A for dimensions). While navigating the maze, mice cannot see the experimenter, which can be a distraction in mazes with transparent walls or open, elevated arms, but they can see distal extramaze visual cues above the maze (see Figure 1). Additionally, this protocol is, in principle, applicable to real-time in vivo imaging and electrophysiology because there is no wiring above the maze, and mice can learn the task efficiently.

Future directions
The current protocol is 15 days long from habituation to the end of the working memory task. We performed the protocol without breaks on weekends; however, having a weekend break after the 6-day habituation should not significantly impact the following 9-day working memory task. Taking a weekend break during the 9-day working memory task may slow down learning in mice. Thus, having two experimenters who are well-trained in handling mice might help reduce experimenters' stress without slowing down the mice's learning. According to Figure 4A, B, the first three days are crucial, as mice achieve more than one perfect trial on average after day four. Therefore, it is advisable to avoid breaks during the initial three days. If the working memory task is too easy for the mice of your interest, increasing the confining time after forced runs could be necessary. These need to be adjusted for individual cases.

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Artery forceps or plier	Any	N/A	For cutting the standard chow
C57BL/6J mice	Jackson laboratory	000664	2~3 months old
Condensed milk (Carnation)	Nestle	N/A	Used as the reward in the task
Distilled water	Any	N/A	For cleaning and diluent of the condensed milk
EthovisionXT13	Noldus	N/A	Control the maze doors and tracking mouse
--Multiple body point module	Noldus	N/A	For Multiple body point tracking
--Trial & Hardware Control Module	Noldus	N/A	For Trial & Hardware Control
-- I/O box	Noldus	N/A	Control the maze doors
Extramaze visual cues	Any	N/A	For mapping the maze by mice
High-resolution camera	Tracksys	N/A	For mouse tracking and filming
Paintbrush	Any	N/A	For cleaning
Paper tunnel	Any	N/A	For handling mice
Paper igloo	Any	N/A	For nesting
Paper nesting stripes	Any	N/A	For nesting
Pipetteman P1000	Gilson	F144059M	For providing diluted condensed milk to the home cage
Pipette tip P1000	Any	Any
Pipetteman P200	Gilson	F144058M	For baiting the maze
Pipette tip P200	Any	Any
Radial arm maze	Tracksys	N/A	Custom made
Room lamps	Any	N/A	For even illumination in the maze
Small plastic cup	Any	N/A	For providing diluted condensed milk to the home cage
Speakers for white noise	Any	N/A	For playing white noise
Standard chow	Any	N/A	For diet restriction
tissue paper	Any	N/A	For cleaning
Verifocal lens	Tracksys	N/A	5-50 mm, manual focus, manual iris
Wooden bite block	Any	N/A	For nesting
50 ml centrifuge tube	Falcon	Falcon™ 352070	For mixing the condense milk with distilled water (1:1)