Many water mazes have been developed, but the one that is referred to as ‘the water maze’ was developed by Richard Morris1
. The maze was designed as a method to assess spatial or place learning and herein will be referred to as the Morris watermaze (MWM). Morris described the basic procedures in 1984 (ref. 2
) and subsequently added details and procedures for assessing related forms of learning and memory3
. Several characteristics have contributed to the prevalent use of the MWM. These include the lack of required pretraining, its high reliability across a wide range of tank configurations and testing procedures, its cross-species utility (rats, mice and humans (in a virtual maze4
)), extensive evidence of its validity as a measure of hippocampally dependent spatial navigation and reference memory5
, its specificity as a measure of place learning, and its relative immunity to motivational differences across a range of experimental treatment effects that are secondary to the central purpose of the task (genetic, pharmacological, nutritional, toxicological and lesion). Although the latter is a general characteristic shared by all water mazes6
, the MWM capitalizes on this strength. For example, hippocampal and septohippocampal lesions in rats reliably induce hyperactivity, but such animals show deficits in the MWM7
. At the opposite pole, treatments that induce hypoactivity can be dissociated from learning deficits in the MWM. For example, it has been shown that MWM learning impairments are independent of locomotor effects because land-based locomotor reductions did not affect swimming speed. Moreover when the experimental animals have deficits during probe trials, this further dissociates learning from performance because measures recorded on probe trials are insensitive to swimming speed8
The use of the MWM in assessing learning and memory has been reviewed9,10
, as has the relationship between performance in the MWM and both neurotransmitter systems and drug effects11
. MWM performance has been linked to long-term potentiation (LTP) and NMDA receptor function12–15
, making it a key technique in the investigation of hippocampal circuitry. In addition, it has been shown that there is involvement of the entorhinal and perirhinal cortices, as well as involvement of the prefrontal cortex, the cingulate cortex, the neostriatum, and perhaps even the cerebellum in a more limited way10
Despite extensive use of the MWM, the task has not always been used optimally. Some of this stems from an under-appreciation for the aspects of the apparatus and testing procedures that are most salient for obtaining the best possible data. Here, we provide a description of the apparatus, its key features, and protocols that are effective and reliable for detecting drug/lesion-induced changes in spatial learning and memory16–20
or changes that arise as a result of genetic manipulations21–28
. We also provide variations to the basic protocol that can be used to enhance assessment of spatial navigation and/or test for related types of learning (latent, discrimination, and cued learning or working memory).
The MWM is not a maze in the usual sense—that is, it is not a labyrinth; rather, it is an open circular pool that is filled approximately half-way with water. The interior is made such that it is as close to being featureless as possible. It is a ‘maze’ in the sense that the animal must search in order to locate a relatively small goal (a hidden platform) that is submerged below the water surface and placed in a fixed location. The platform is camouflaged either by placing opacifying materials in the water (typically, tempera paint or polypropylene pellets), by creating a nearly invisible platform-to-background color match, or by using transparent platforms against a colored background, thereby making it indistinct given the low visual aspect ratio to the water as seen by the animal when swimming.
It is standard to designate two principal axes of the maze, each line bisecting the maze perpendicular to one another to create an imaginary ‘+’. The end of each line demarcates four cardinal points: North (N), South (S), East (E) and West (W). These are not true magnetic compass directions but refer to S being the experimenter’s position, N being at the opposite point, E being to the experimenter’s right and W being to the experimenter’s left. Dividing the maze this way creates four equal quadrants. The platform is positioned in the middle of one of the quadrants. One can either keep the platform in one quadrant for all trials or test one-quarter of the animals with the platform in each of the quadrants. The latter approach counterbalances for possible quadrant effects. One can even use eight different platform positions22
. The platform is usually located half-way between the center and the wall, regardless of the quadrant selected, although other arrangements are sometimes used29
Place or spatial learning is the most basic MWM procedure. The concept behind it is that the animal must learn to use distal cues to navigate a direct path to the hidden platform when started from different, random locations around the perimeter of the tank. If there are no proximal cues available, the use of distal cues provides the most effective strategy to accomplish this. Most protocols use four start locations: N, S, E and W. Animals are given a series of daily trials using a random or semi-random set of start locations. Semi-random start position sets are most common, such that the four positions are used, with the restriction that one trial each day is from each of the four positions. A few investigators use eight start locations30
. One concern about the cardinal start positions is that they are not equidistant from the goal, creating short and long paths to the goal. Even in a large maze, a rat starting at E, with the goal located at SE, has a short path to the goal. There is no perfect solution to this problem. A partial solution that we have used is to use only distal start locations18
. By this, we mean that if the goal is SE, then one can use start locations of N, W, NE and SW. Although not equidistant from the goal, these start positions are closer to being equal in length than using start positions that are adjacent to the goal. Another approach might be to use only two start positions, such as N and Wonly, but one must then be concerned that animals might memorize specific routes rather than use distal cues. A third approach is to use three start positions, each in quadrants other than the one containing the platform31
, however only two of these are equal in length.
illustrates a set of semi-randomly selected distal start positions for basic acquisition training, with the platform being located in the SW quadrant. These are designed so that the animal is not able to learn a specific order of right or left turns to locate the platform, while using each of the four start positions once each day. As can be seen in , the learning trials are conducted over 5 days, with 4 trials per day. The interval between trials can vary from 10–15 s to 5–15 min. If an animal fails to find the platform within the allotted time, it is usually picked up and placed on the platform for ~15 s, although some prefer to guide the animal to the goal based on evidence that it is the middle portion of the swim path that seems to be most important in learning how to navigate to the goal32
Morris water maze spatial (hidden platform) start positions.
To assess reference memory at the end of learning, a probe (transfer) trial is given. The most common method is to administer one probe trial 24 h after the last acquisition day. With some procedures, the probe trial is administered immediately following the last learning trial; however, this cannot differentiate between short- and long-term memory, as it may reflect memory for the most recent training session. A long interval between the last training trial and the probe trial is essential if reference memory is to be determined independent of the memory of the last training session.
Additional probe trials are sometimes interspersed during the learning phase: these are often given before the first learning trial of the day. These additional probe trials may help to determine the rate of memory consolidation, as this allows the gradual emergence of goal quadrant preference to be seen across days. However, caution should be exercised not to conduct too many probe trials as these are extinction trials and may slow the rate of learning.
It is increasingly common and frequently informative to relocate the platform to another quadrant (usually the opposite one) and administer another set of four trials per day for 5 additional days (). This is often called reversal learning, although the term is not precise, as swimming to an opposite quadrant is not the mirror image of the initial problem as it is in a T-maze. Reversal learning in the MWM reveals whether or not animals can extinguish their initial learning of the platform’s position and acquire a direct path to the new goal position. Tracking patterns typically reveal that mice swim to the previous location first, then begin to search in an arching pattern to reach the new goal (). Even after multiple trials, mice do not entirely abandon their initial learning strategy and begin trials by starting to move towards the original platform position, then turn and swim more directly to the new goal. Rats, on the other hand, rapidly switch their search strategies to the new goal position (). In fact, rats switch away from the old goal location so rapidly that return visits to the original platform location above chance (i.e., 25%) cannot be seen in the average of the first four trials on reversal day 1 but may be seen on individual trials within the first day of reversal testing. As in the acquisition phase, at the end of the reversal phase, a reversal probe trial is given 24 h later.
Figure 1 Percent time in each quadrant of Morris water maze performance on each day of testing in C57BL mice. The results were averaged across four trials per day (mean ± s.e.m.) in untreated C57BL male mice during the reversal phase of learning — (more ...)
Figure 2 Percent time in each quadrant of Morris water maze performance on each day of testing in Sprague–Dawley rats. The results were averaged across four trials per day (mean ± SEM) in untreated Sprague–Dawley male rats during the reversal (more ...) Spatial double-reversal with a smaller platform
Many variations can be added to the basic MWM procedures and these can add valuable information for understanding the deficits that are observed or may even unmask more subtle deficits that are not seen during acquisition or reversal learning. One procedure that has been effective in our hands has been to move the platform again, either back to the original goal (double-reversal) or to a different quadrant (shift), but with an additional change: use of a smaller platform17
. For example, if the starting platform is 10 × 10 cm, the reduced platform may be 5 × 5 cm. This reduction in platform size taxes the spatial accuracy requirements of the animal and has revealed the effects of some drugs or doses that are not seen during acquisition or reversal16,20
. A reduced platform probe trial is also given 24 h after the end of this phase of testing.
Another procedure is to conduct a set of reversal or shift phases serially19
. This allows an examination of the animals’ flexibility in their ability to learn across multiple phases of new learning. The data also demonstrate the effects of moving the platform to different quadrants. For example, if the platform is shifted to an adjacent quadrant, new learning is more rapid than if shifted to an opposite quadrant19
Spatial working memory
The procedures described above are for the assessment of trial-independent learning (that is, the goal does not move from trial to trial during a given phase of testing). To assess working or trial-dependent learning and memory, a different method is required. In this procedure, which is also called matching- to-sample, the platform is relocated every day and the animal is given two trials (or more) per day (see ). On each day, the first trial represents a sample trial. During the sample trial, the animal must learn the new location of the platform by trial-and-error. Trial 2 (or any successive trial) is the test or matching trial in which savings in recall between Trial 1 and Trial 2 are measured. Trial 2 begins after a 15-s inter-trial interval. If the animal recalls the sample trial, it will swim a shorter path to the goal on the second trial. As the platform is moved daily, no learning of platform position from the previous day can be transferred to the next day’s problem; hence, recall on each day during Trial 2 is dependent on that day’s sample trial and measures only temporary or working memory.
Sequence of start and goal positions for assessing trial-dependent (working) spatial learning and memory.
The MWM can also be used to assess visual discrimination learning3
. In this procedure, two visible platforms are used that are distinct from one another such as one being white and one being black. One is the standard fixed platform that is raised above the water and the other platform floats from a tether. The task for the subject is to learn which platform can be used for escape from the water and which cannot. The accuracy of the animal’s choice across successive trials is an index of its ability to differentiate the stimulus information of the ‘true’ goal relative to the ‘false’ goal.
In latent learning, the idea is to place the animal on the platform before each trial rather than after. This will allow one to determine how much of the spatial learning stems from navigating to the platform compared with orientation to the goal once there. Morris has described this procedure elsewhere3
A control condition that is frequently used in the MWM is to test the animals for their ability to learn to swim to a cued goal. In this procedure, curtains are closed around the maze to reduce the availability of distal cues. The curtains interfere with the animal’s access to distal cues that could be used to spatially navigate. The platform is the same as in the hidden platform version, except that it is either elevated above the water surface19
or is kept submerged but a ‘flag’ is mounted that extends above the water surface by approximately 12 cm (ref. 33
). Although both methods work, we find that the version with a flag seems to be more efficient, as it is readily recognizable from across the pool, whereas the protruding platform may not be. This ‘cue’ is designed to allow the animal a direct line-of-sight to the platform’s location. To ensure that the animal is using this proximal cue to locate the platform, the location of the goal and the start are both moved to new positions during each trial (). In this way, the subject cannot use distal cues to solve the problem. The only cue that reliably indicates the location of the platform relative to the start is the cue that is attached to the platform. Morris introduced this as a control procedure as part of his original description of the test1
. Unfortunately, this procedure is all too often omitted, yet its value is unmistakable. If subjects are impaired in cued learning, there is a potentially serious concern about whether a spatial deficit is present. This is because cued learning requires the same basic abilities (intact eyesight, motoric ability (swimming), basic strategies (learning to swim away from the wall, learning to climb on the platform)) and the same motivation (escape from water) as the spatial version of the task. Therefore, if the subject cannot perform the cued task, doubt is cast on its capacity to learn using distal cues in the spatial version. This task can be administered before or after the spatial version, but administering it before has advantages, especially for mice. Some animals find the platform, but then jump back into the water and continue searching. Presumably, this reflects an effort to find another route of escape. In the first few trials, some animals are sufficiently activated by being in the water that it is not always clear that they recognize that the platform is an escape when they first locate it. Therefore, testing animals first in cued trials eliminates the problem of animals not acquiring the appropriate subordinate skills before they are presented with the spatial version of the task.
A cued learning trial pattern.
Cued learning is basically a control procedure, but it is not the only one available. For example, Cain34–36
has shown that some drugs interfere with sensorimotor function and this can interfere with the animal’s ability to recognize that the platform is the goal. He has proposed several ways of determining whether sensorimotor interference is occurring. He suggests measuring thigmotaxis, or the tendency to cling or follow the wall around the outer perimeter of the tank, as one index to reflect that the animal is not problem-solving. Excessive thigmotaxis (especially in rats) indicates that the animal is not focusing on the task appropriately because one of the first things that animals have to learn is that there is no escape located around the perimeter of the tank. Having learned this, most animals swim away from the wall and then, by weaving or looping search patterns, eventually find the platform. Not learning this basic approach indicates that the animal may not have adequate awareness of its surroundings. Other measures of impaired sensorimotor interference are excessive jump-offs, swimovers and/or deflections. Rats that reach the platform but do not climb on it, or do not stay on it, are not acquiring the requisite association between the platform and escape. Some of these behaviors may be seen during early trials even in control animals, but these usually disappear within a day. If such behaviors are more frequent in the experimental group, however, questions should be raised about whether spatial learning can be satisfactorily assessed.
Cain has suggested that one way to solve the problem of sensorimotor interference is to compare two groups of experimental animals: one tested in the standard procedure and one pretrained using a ‘non-spatial’ training procedure. The pre-trained group is given a series of hidden platform trials in which the start and goal are moved randomly on every trial, as is done in cued learning, but here the curtains are left open and the platform is hidden. The task cannot be solved using spatial navigation because of the randomized start-goal combinations but it teaches the subject the basic task requirements — namely, the escape can only be found by searching, the goal is located somewhere away from the wall and the platform is the goal. Cain34–39
have demonstrated that ‘non-spatial’ pretraining can separate components of learning that are not spatial from those that are and this, in turn, can change the interpretation of the findings. As non-spatial or strategy pretraining and cued training both have the effect of teaching animals the basic task requirements and tend to eliminate behaviors such as swimovers, jump-offs and even diving, it can be helpful to conduct cued trials first. Strategy pre-training is not usually necessary unless the data show that thigmotaxis or platform recognition behaviors indicate that sensorimotor problems are present. Acute pharmacological studies, however, may need non-spatial pretraining to ensure that non-cognitive effects are not interfering with maze performance, whereas delayed or long-term drug effect studies may not need this extra procedure.
Another approach to determine whether or not animals have any underlying sensorimotor deficits is to assess swimming speed. Often this is done in the maze during learning trials41
. Alternatively, one can pre-test animals in a separate apparatus, such as a long, straight swimming channel. We use a 15 × 244 cm water-filled channel with an escape ladder or platform at one end33
, although shorter channels have also been used31
. This task requires virtually no searching, hence virtually no learning. The first one or two trials serve to acclimate them to swimming and the rats quickly recognize that the escape can be found by simply swimming from one end to the other. During subsequent trials (a total of four is typical), rats swim as fast as they can to get from the start to the goal. This provides a measure of basic swimming ability and motivation to escape from water, and can be used to determine whether or not animals are motorically and motivationally equivalent across groups prior to MWM trials. An analysis of the average of these trials or use of the fastest trial provides assurance that MWM trials can be interpreted correctly.
Summary of the MWM
Spatial mapping versus working memory hypotheses and the data supporting each using the MWM has been reviewed in detail elsewhere5,42
. There are many tests that have been used to assess these functions, of which the MWM is but one. However, the MWM has become an important, even dominant, method. As with all methods, the MWM has strengths and weaknesses; however, most of its perceived weaknesses arise from the use of mazes that are too small, protocols that do not adequately assess learning, failure to provide an appropriate interval between training and probe trials to assess reference memory, or lack control procedures to assess non-spatial factors. Despite this, the MWM has become more widely used than its predecessors (radial-arm maze, passive avoidance, T-mazes and their variations) since its introduction 25 years ago. This increased utilization arises because the effects on MWM performance after treatment have been more widely replicated than the effects observed with any other learning task, and the MWM is relatively straightforward to set up. There can be little doubt that the MWM has significantly advanced our understanding of the relationship between NMDA receptors, synaptic plasticity and learning43
, and it continues to be used in new applications for the assessment of other types of learning. As use of the task has grown, so too have the number of methodological variations, some of which have extended its utility. The protocols described here provide guidance that can help users avoid the most common pitfalls.
The MWM is primarily a test of spatial learning and reference memory and that remains its principal strength. Detailed analyses have shown that rats can solve the task using a minimal set of cues that involve angular separation and distance from the tank wall29
. Such data show that when properly configured and utilized, with the inclusion of appropriate control procedures, the MWM is a powerful technique for assessing spatial mapping. Appropriate modification of the basic protocol makes it a flexible tool that can be applied to probe spatial learning in more depth or to assess other forms of learning and memory.