Learning spatial orientation tasks in the radial-maze and structural variation in the hippocampus in inbred mice
© Crusio and Schwegler. 2005
Received: 07 February 2005
Accepted: 22 April 2005
Published: 22 April 2005
In the present paper we review a series of experiments showing that heritable variations in the size of the hippocampal intra- and infrapyramidal mossy fiber (IIPMF) terminal fields correlate with performance in spatial, but not non-spatial radial-maze tasks. Experimental manipulation of the size of this projection by means of early postnatal hyperthyroidism produces the effects predicted from the correlations obtained with inbred mouse strains. Although the physiological mechanisms behind these correlations are unknown as yet, several lines of evidence indicate that these correlations are causal.
Several learning tasks are available to test spatial orientation abilities in mice. The most widely applied one is probably the water navigation task (also known as the "Morris maze"), which was developed originally for rats . However, it has been noted that mice are animals living in dry habitats [2, 3], so that a swimming task may be less appropriate to them, because of the stress it may be expected to induce [4, 5]. A factor analysis of data from several thousand mice was carried out by Wolfer and Lipp , who reported that only the third and least important factor showed loadings of behavioral variables related to spatial orientation, explaining less than 20% of the observed variance in behavior. Indeed, mice with hippocampal lesions still can improve their performances in this task over time . This does not necessarily mean that we should abandon the Morris water navigation task for use with mice, it just means that this test may reveal differences between groups of animals that may relate to different factors and therefore should not automatically be interpreted as differences in spatial learning ability. It would therefore appear that appropriately designed dry-land mazes might assess spatial learning capacities of mice more specifically than water mazes. Among the available mazes, the radial maze appears to be especially suitable .
Designing a radial maze
Heritable neuroanatomical variation in the hippocampus
Simple tasks with all arms baited
At first sight, this result seemed to confirm our hypothesis: animals with larger IIPMF projections committed fewer errors, mastering the task more rapidly than animals with smaller IIPMF. However, matters were perhaps more complicated than that. Upon closer examination, it appeared that many animals used a kinesthetic strategy to solve the task, visiting adjacent arms in a clockwise or counter-clockwise fashion . Whether such a strategy is based on spatial orientation capabilities or not, is not directly evident. Therefore, we decided to modify the radial maze task, as it was known from work with rats that confining subjects for 5 sec to the central platform in between subsequent arm choices interrupts this kind of chaining response . In addition, observations made during behavioral testing suggested that mice are probably not using extra-maze cues when opaque arms are used as they rarely seemed to look upwards. We therefore replaced the PVC arms with arms made of clear Plexiglas and installed guillotine doors to enable the application of a confinement procedure. To facilitate the use of extra-maze cues for spatial orientation, we placed several objects close to the maze (of course, the experimentator is already a very visible cue in him/herself). This was done because the best-performing strain in the pilot experiment, C3H/HeJ, carries the Pde6b rd1 mutation causing retinal degeneration . Although these animals are not yet blind at the age that we use them (about 3 months, see [20–22]), their visual acuity obviously will be severely impaired, so we wanted to make this task as easy on their visual systems as possible.
At the same time, we tested the same number of animals and strains in another radial-maze task that did not require any spatial orientation in order to be solved . Here, opaque PVC arms were used and instead of manually operated guillotine doors we employed perforated aluminum plates fixed to the floor with adhesive tape. Subjects could easily open the doors, but as this took a few seconds, this procedure was also expected to disrupt any kinesthetic strategies. As with the previous spatial task, performance in this experiment ranged widely between different inbred strains and kinesthetic strategies were, indeed, absent. However, no correlation whatsoever with hippocampal mossy fibers became apparent (data not shown, see ).
The results obtained were in accordance with our hypothesis that sizes of the IIPMF would correlate with spatial learning capacities, but not with nonspatial learning abilities. These data therefore provided support for the cognitive mapping theory of O'Keefe and Nadel , which postulates that the hippocampus is uniquely involved in the regulation of spatial, allocentric memory. However, an alternative explanation was available, too. Olton  has hypothesized that the hippocampus regulates working memory, as opposed to other brain systems that would modulate reference memory, regardless of whether the information concerned was spatial or nonspatial in nature. Under this hypothesis, working memory stores information that is pertinent to one trial only (for instance, which arms have already been visited), but which has to be erased before the next trial to allow correct performance. Reference memory concerns information that is pertinent to all trials (for instance, the fact that food can be found at the end or an arm). Obviously, our spatial task had been a working memory task, whereas the nonspatial task was a reference memory task. Our results were therefore compatible with both competing theories, that of O'Keefe and Nadel  and that of Olton . We therefore modified our task yet again, to allow simultaneous measurement of working and reference memory in both spatial and nonspatial versions of the radial maze.
More complex tasks dissociating working and reference memory
Following Nadel and McDonald , we trained animals from the same nine inbred strains on a task in which only four out of the eight arms were systematically rewarded, the other four arms never containing any accessible food . Two experiments were done. In one the task was spatial, using the radial maze with Plexiglas arms and guillotine doors as described above. In the other one, the task to be mastered was non-spatial, using the radial maze with opaque PVC arms, combined with guillotine doors. In both tasks, animals were confined to the central platform for 5 sec between subsequent arms choices. In the spatial version, mice were trained to locate four food rewards that were always placed in the same set of four arms. Each individual mouse had its own set of four rewarded arms. Following Olton's definition, entries into an arm that is never baited constitute a reference memory (RM) error, whereas repeat entries into an arm that has been visited previously constitute working memory (WM) errors. To prevent animals from using within-maze cues, the maze was rotated 45° at the end of each day (between subsequent trials), so that intra-maze and extra-maze cues were dissociated. This procedure prevented animals, e.g., from following hypothetical olfactory trails and forced them to use extramaze cues exclusively. In the nonspatial version, arms were marked by different black-white patterns on their floors. Food rewards were now associated with different sets of black/white patterns, each individual mouse again having its own combination of rewarded and non-rewarded patterns. As in the spatial task, RM and WM errors can now be defined.
Using the radial maze to demonstrate mutational and pharmacological effects
Of course, correlations between two variables need not indicate a causal relationship and the IIPMF-spatial learning correlation might be spurious. Hypothetically, a third, as yet unknown, neuronal variable might be the one causing the observed strain differences in learning. The IIPMF-learning correlation would then only appear because this hypothetical third variable itself would be correlated with the IIPMF. However, in the present case we believe that there are strong indications that this correlation is, indeed, causal. First, there is the remarkable consistency and strength of the correlations reported. If a third variable would be directly correlated with learning performance, the IIPMF correlation would be only secondary and the third variable would have to correlate with learning even stronger than the IIPMF do. This would be difficult to imagine. Second, a correlation between strain means differs in one important respect from ordinary correlations, estimated from individual values. Namely, such a correlation represents a genetic correlation, meaning that gene effects on one variable are correlated with gene effects on the other variable [34, 35]. Such a situation makes it highly likely that the statistical relationship found is, indeed, a causal one. Finally, we have also addressed this question in a pair of experiments in which newborn pups of a strain (DBA/2) known to possess scant IIPMF projections and feeble learning capacities in the radial maze were treated with thyroxin in the early postnatal period [36–38]. This treatment induces an increase in the size of the IIPMF in adults and we found that this increase was accompanied with a significant improvement in the spatial learning capabilities of these animals, both in a task in which all arms were baited as well as in a task in which only 4 of the arms were consistently baited.
Other behaviors correlated with hippocampal neuroanatomy
Although the present review concerns radial-maze learning, we would like to briefly mention some other behaviors that have been found to correlate with the IIPMF. The very first correlation that was reported, with two-way active avoidance learning, has already been mentioned above (for a review, see also . Other correlations that have been found are with intermale aggression [40–42], paw preference , reversal learning in a water navigation task [44, 45], visual and tactile discrimination in a Y maze , and exploration [47–49] and habituation  in an open field.
Taken together, we conclude that in inbred mice the hippocampal intra- and infrapyramidal mossy fiber projection plays an important role in the regulation and/or modulation of spatial orientation capabilities in the radial maze. Larger IIPMF projections go with better learning capabilities in the spatial radial-maze tasks described above. These correlations are specific to spatial learning, as no correlations were found in radial-maze tasks that could be solved using non-spatial cues or that employed elevated mazes, possibly inducing increased levels of anxiety in the experimental subjects.
At this point it is not yet fully understood why variations in the size of the IIPMF have such drastic consequences for an animal's behavior. It has been shown that these variations are associated with differences in spontaneous bursting in region CA3  and with differences in LTP . LTP is generally regarded as the most promising physiological mechanism underlying learning and memory, although the extent of its implication in these processes remains controversial (see  for a critical discussion). However, it was recently found that blocking of mossy fiber LTP or LTD does not abolish spatial learning capabilities in mice . Therefore, the IIPMF apparently modulate behavior by other mechanisms. One possibility may be that larger IIPMF would somehow facilitate LTP in the hippocampal CA1 region. This hypothesis would be consistent with our findings that larger IIPMF go with better spatial learning abilities and diminished two-way active avoidance learning, behaviors that are abolished, respectively enhanced, after a hippocampal lesion . Further research is clearly needed to address these questions.
Finally, among the different hippocampus-dependent radial maze tasks presented above, the simple spatial one (all eight arms rewarded, short confinement to the central platform between subsequent arm choices) appears to be the most useful task: It is rapid (in our protocol we use 5 daily trials only, each trial taking on a few minutes near the end of training) and we have used it successfully to investigate effects of pharmacological treatments [55, 56] or to detect subtle mutational effects .
We would like to thank our colleagues and students who over the years have collaborated with us on the studies presented above: Ingrid Brust (Heidelberg, Germany), Pascale Guillot (London, UK), Laure Jamot (Marseille, France), Hans-Peter Lipp (Zurich, Switzerland), Charlotte Marican (Paris, France), and Frans Sluyter (London, U.K.).
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.